Prospective crime mapping in operational context Final report

1. UCL JILL DANDO INSTITUTE OF CRIMESCIENCEProspective crime mapping in operationalcontextFinal reportShane D JohnsonDaniel J BirksLindsay McLaughlinKate J BowersKen PeaseUCL, Jill Dando Institute of Crime ScienceSecond Floor, Brooke House2-16 Torrington Place, LondonWC1E 7HN Tel. (020) 7679 0809Fax. (020) 7679 0828Email: shane.johnson@ucl.ac.ukOnline Report 19/07

AcknowledgementsThe authors are grateful to a number of people for their help and advice throughout thisproject. These include but are not limited to Assistant Chief Constable John Wright, DCI RickGooch and other members of the Command Team, Derbyshire ‘A’ Division. IntelligenceAnalysts Deborah Rimell and Bill Wallage, PC Hayley Vincent for filling out the tactical optionslog; Sergeant Kevin Pellatt and David Lynam (partnership analyst) from Safer DerbyshireResearch and Information Team who provided the data analysed; Inspector Matt Thompson‘A’ Division Community Safety; All ‘A’ Division front-line officers and Sergeants; Sgt AlanBeeson for driving Lindsay around to all the sections on the day of the surveys; SteveBrookes and Phil Taylor from Government Office for the East Midlands; and, MichaelWilkinson, Lindsey Poole, Mark Bangs, Steve Wilkes and Niall Hamilton-Smith from the HomeOffice. We would also like to thank two anonymous reviewers for their insightful comments.i

ContentsAcknowledgementsExecutive summaryiv1. Introduction 1Background and past research 1Aims and objectives 32. Testing the generalisability of prospective mapping 4Is the risk of burglary communicable in the East Midlands? 6Predicting the future 183. Tactical options and selecting a pilot site 28Potential pilot sites 28A tactical options matrix for reducing burglary 304. System development and evolution 38Time of day consistency? 39Conclusion 415. Process evaluation 43Process evaluation methodology 43‘A’ Division, management and day-to-day running 44IT and dissemination 45Tactical delivery 52Summary 576. Changes in patterns of burglary 59Change in the time of day burglaries were committed 627. Conclusions 67References 70Appendices1 The information technology nexus 742 Prospective Mapping Survey 793 Detailed evaluation methodology 834 Promap graphical user interface and an illustration (step by step) of how thesystem is used 95List of tables2.1 Knox contingency table example 42.2 Knox ratios for Mansfield 72.3 Monte-Carlo results for Mansfield 72.4 Weekly Knox ratios for Mansfield 82.5 Knox ratios for Wellingborough 82.6 Monte-Carlo results for Wellingborough 92.7 Weekly Knox ratios for Wellingborough 9ii

2.8 Knox ratios for Ashfield 102.9 Monte-Carlo results for Ashfield 102.10 Weekly Knox ratios for Ashfield 112.11 Knox ratios for Corby 112.12 Monte-Carlo results for Corby 122.13 Knox ratios for ‘A’ Division 122.14 Monte-Carlo results for ‘A’ Division 132.15 Weekly Knox ratios for ‘A’ Division 132.16 Summary of the analyses concerned with the communicability of risk 142.17 Weekly Knox analysis for area 1 152.18 Weekly Knox analysis for area 2 162.19 Weekly Knox analysis for area 3 162.20 Weekly Knox analysis for area 4 162.21 Weekly Knox analysis for area 5 172.22 Average number of crimes correctly identified per forecast for cumulative methods232.23 Average percentage of crimes correctly identified per forecast 232.24 Average number of crimes correctly identified per forecast for single point methods242.25 Average percentage of crimes correctly identified per forecast 242.26 Predictive accuracy for analyses for which the same number of cells wereidentified by each method 253.1 Comparison of three potential pilot sites 293.2 Tactical options matrix 324.1 Accuracy of the prospective model including the opportunity surface 385.1 Number of times prospective maps were used in ‘A’ Division’s daily briefing 525.2 Sample characteristics 535.3 Number of respondents who had heard of prospective mapping, by section 545.4 Number of times maps were used for targeted police activity 545.5 Number of respondents who were either involved in or responsible foremploying operational tactics, by section 555.6 The interpretation and usefulness of prospective maps 566.1 Change in the volume of burglary and odds-ratio statistics 62A 3.1 Change in the volume of burglary and odds-ratio statistics 85List of figures2.1 The five policing areas in ‘A’ Division 152.2 Two-dimensional and three-dimensional hotspot lattices 192.3 Opportunity surface for ‘A’ Division 212.4 Differences in cells identified as being at the highest future risk byretrospective and prospective methods 25iii

2.5 Illustration of a simple nearest neighbour analysis for two data sets 262.6 Nearest neighbour index: retrospective and prospective methods 274.1 Similarity in time of day for near-repeats and unrelated burglaries 404.2 An example image of the final GUI 415.1 Promap dissemination process across ‘A’ Division 485.2 A series of predictions for one area 495.3 Timeline for prospective mapping pilot in ‘A’ Division 516.1 Time-series graph of the count of burglary before and during pilot 616.2 Changes in the proportion of burglaries committed during the evening overtime (for events for which the interval between the earliest and latest reportingtimes was less than eight hours) 636.3 Changes in the proportion of burglaries committed during the morning overtime (for events for which the interval between the earliest and latest reportingtimes was less than eight hours) 646.4 Changes in the proportion of burglaries committed during the daytime overtime (for events for which the interval between the earliest and latest reportingtimes was less than eight hours) 656.5 Changes in the proportion of burglaries committed during the evening overtime (for events for which the interval between the earliest and latest reportingtimes was less than eight hours) 66A1.1 Internal application utilising existing GIS for visualisation 75A1.2 Stand-alone application utilising existing GIS 76A1.3 Stand-alone application with integrated GIS 77A3.1 Changes in the spatial distribution of risk following the introduction of the pilot 88A3.2 Lorenz curves showing the distribution of burglary risk 90A3.3 An illustration of a triple (bottom) and quad chain (top) 91A3.4 The proportion of events belonging to different k-event series before andduring the pilot 93A5.2 An enlargement of the shift analysis options 96A5.3 An example of fictitious prospective map 96A5.4 Map navigational options 97A5.5 Prospective map magnified to neighbourhood level 97A5.6 Prospective map magnified to street level 98A5.7 Prospective map magnified to household level 98iv

Executive summaryThe systematic identification and management of risk is one element in the Home Office’sreform plan published on 19 July 2006, with a stated emphasis on proactivity in riskmanagement. The research reported here provides an innovative means of doing preciselythat in respect of one offence, describing and applying a technique whose extension to alloffence types would make a significant contribution to realisation of the Home Office’s reformagenda.Knowing when and where to deploy resources is pivotal to the crime reduction enterprise.The attempt to identify and police ‘hotspots’ of crime has for some time been a part of crimereduction strategy. The positive effects of such measures have been acknowledged, with arange of interventions based on this approach having significant impacts on levels of crime.The aim of the project reported here is to understand the regularities in patterns of burglaryacross a range of geographical areas and to develop and test an emerging forecastingtechnique, prospective mapping, thereby helping the police and their crime reduction partnersto prevent and detect more crime.The risk of burglary is not evenly distributed – some areas experience more burglary thanothers. Within areas, some homes are victimised more than the rest. However, little researchhas focused on the accurate prediction of which areas and locations are most likely toexperience burglary next. Research concerned with crime mapping has focused almostexclusively on the description of what happened last week, last month or over the last year,with the implicit assumption that the future will be much like the past. Insofar as crime moves(and it does), such an approach is at best sub-optimal. The current study sought to developan accurate way of forecasting where burglary is most likely to next occur, and to decidewhether the resulting system had potential for use in operational policing. In addition toaddressing the technical issues of how to forecast future patterns of burglary, the authorsattempted to identify issues of implementation that might impede police adoption of predictivemapping systems and how such mapping might be integrated with other approaches to crimereductionSpatial and temporal patterns of crime are fluid. Research by the authors and others haddemonstrated that the risk of burglary appears not only to move, but to cluster in space andtime in much the same way as a communicable disease. When a burglary occurs at onehome, another is likely to occur swiftly nearby. As time elapses, this risk decays so that afterfour to eight weeks, homes located near to a previously victimised home experience only alevel of risk normal for the area in which they are located. If such patterns are ubiquitous (anotion supported by the work reported), the risk of burglary moves. The consequence is thatthe location of future events might be better predicted by means more sophisticated than thesimple extrapolation of past patterns.The central aims of the project were as follows.• To determine whether patterns of burglary were communicable across diverse areasof the East Midlands, and if so, whether the pattern varied between areas.• To develop a predictive mapping system usable in an operational policing context.• To test the accuracy of the system and compare it with contending alternatives.• To determine how the system could be used operationally, identify obstacles toimplementation, establish how it was received by those who might use it, and identifynecessary refinements.• To provide an idea of the likely efficacy of the system during a field trial by evaluatingits impact on crime and influence on crime reduction strategies in the area.v

Project outcomesPatterns of burglaryUsing a modification of an approach developed in the field of epidemiology, analyses wereconducted to see if burglary does cluster in space and time and if the patterns vary acrosslocation. Briefly, to do this, for each area each burglary event is compared to every other andthe space-time distance between them recorded. The pattern of results observed is thencompared with what would be expected on the basis of chance, determined using what isknown as a Monte-Carlo simulation. Burglary was considered to be communicable if moreevents occurred near to each other in both space and time than would be expected on thebasis of chance.Without exception in the East Midlands areas studied, the risk of burglary is communicable upto a distance of around 400m for at least one month. This finding was used as a general rulein the construction of basic prospective maps. Sensitivity analyses were conducted toexamine the duration of the elevation in risk. These suggested that risk (albeit diminishing)extended beyond one month and in most cases up to around eight weeks.The patterns were to some extent specific to the time of day considered. For example, if aburglary occurred at one location during the afternoon, a further burglary was more likelynearby soon after and at a similar time of day. In supplementary work not reported here, theftfrom (but not theft of) vehicles was also found to be communicable.Developing the predictive system and measuring its accuracyOn the basis of the above findings, a method of prospective mapping (hereafter, Promap),based on the authors’ previous work, was developed. Across the different areas studied inthe initial phase of the research, police analysts were already using descriptive hotspotmapping. Promap was tested against an optimised version of the ‘retrospective’ hotspottingmethod in current use. The optimised versions were substantially more predictive than mapswhich resemble those typically used in crime analysis. Promap outperformed the optimised‘retrospective’ hotspot mapping system. It did this in three ways.• Promap correctly predicted more burglaries than other methods. For example, the finalversion of Promap could identify the locations of 78 per cent burglaries that occurredwithin the next seven days of a forecast, whereas for the same interval theretrospective model could identify only 51 per cent.• Relative to the retrospective maps, it yielded hotspots which formed more solid,coalescent, hence readily patrollable, areas.• The final version of Promap accurately predicted more crime while identifying a smallerarea than other methods. For example, the same fraction of burglaries occurring withintwo to seven days of a prediction could be forecasted by identifying patrolling areas halfthe size of those generated by retrospective mapsImplications for operational policingResearch conducted by the authors and others suggests that police officers’ perceptions ofwhere burglary hotspots form are often imperfect. This is particularly true for recent ratherthan enduring problems. The fluidity of burglary risk provides an explanation for this. Crimemoves and patterns evolve. Why should police officers be able to anticipate such changes?They should not, but computerised support systems that can assist in the crime reductionenterprise should.The advantages of Promap are that it could act to facilitate more targeted crime reductioninterventions, increasing the likelihood that resources are deployed to the right places at theright time, rather than where they were needed last week or last month (as with conventionalhotspot methods). Risky areas can be better defined so that patrols could move throughpriority areas more efficiently, spending less time in places with lower risk. The maps can alsobe produced for specific shifts to ensure their continued relevance over the course of the day.vi

Developing predictive mapping in an operational policing contextSupport systems are only valuable if they can be used and understood by those operatingthem. Promap software was developed for use in one police Basic Command Unit (BCU) inthe East Midlands in consultation with local police and community safety practitioners. Usingthe software, maps could easily be produced at regular intervals by crime analysts, discussedduring shift briefings and provided to beat officers. The maps clearly defined the areas withthe highest predicted risks, against a background of the housing distribution and significantgeographical points of reference, which could then be used as guides to patrolling. Inresponse to feedback, different maps were generated for each of the three police shifts of theday.The system was modified in response to practitioner requests. Considerable effort wasexpended to optimise the algorithms used to ensure that the system could generate mapsrapidly. The final version could generate maps for the entire participating BCU in around 20seconds. A different system which generates descriptive (i.e. not predictive) maps, took tenminutes to complete an analysis of the same area, with extra time required to display theresulting output.Issues of implementation encountered in situFollowing consultation with the staff in the pilot BCU and their crime reduction partners, thesystem was modified and used in an operational context over a period of six months.Promaps were generated for each of five sections which comprised the BCU by crimeanalysts located at police headquarters, and disseminated using the force IT system.A process evaluation was conducted over the implementation period to see how the systemwas used. This involved observation of briefing meetings, a log of the tactical options used inresponse to the maps, and a survey of front-line police officers.Issues with system implementation• Despite the fact that the system itself was considered simple to use, changes of keypersonnel (including the BCU commander) and force IT requirements (which initiallyrendered the system unnecessarily complex) made for a slow start. Those surveyedcame to regard it as useful to the point of enquiring about the possibility of its extensionto cover other crime types.• Timely dissemination of relevant maps to front-line officers was a source of initialdifficulty, resolved to general satisfaction distressingly late in the pilot period. At firstthere were issues with the physical transfer of maps from the crime analysts, located atpolice headquarters, to the staff throughout the BCU. The maps were initially generateddaily but beat officers felt that there were only minor changes in the maps whengenerated with this frequency. Eventually, these issues were partially resolved byproviding local staff access to the mapping software, and by producing the maps twicea week. Implementation issues of this kind currently represent one of the mostimportant limiting factors in prospective mapping utility, but their resolution is mostly amatter of ensuring that basic IT infrastructure is adequately configured, and the systemis not hobbled by adherence to IT custom and practice.• Part of this project involved a review of possible tactical options that could be used inconjunction with the new maps. These ranged from well-established anti-burglaryinitiatives, such as target hardening and police patrols to novel techniques suggested inthe light of the burglary patterns found. The review of each technique includeddocumented success or failure, financial costs, and the speed with whichimplementation was plausible, swiftness of implementation being important in thecurrent project.It appeared that the most favoured methods were those that combined the maps withother local intelligence (such as data on known offenders) to direct police patrols, andvii

those that involved collaboration with other crime prevention partners. The point ofcentral importance is that the availability of accurate prospective maps requiresreconsideration of tactical options in order for their potential to be realised.• Interviews with front-line officers were undertaken towards the end of the project.Across the five sections of the BCU, there were differences in the degree to whichofficers could define what the system actually did. In one area, only 27 per cent ofofficers interviewed could provide an accurate description, although in the remainingfour areas a better understanding was apparent, with 75-92 per cent providing gooddefinitions. There were also differences across the five areas with respect to howfrequently the maps were reported to have been used to inform police tactics.Typically, the maps were reported to have been used more often in those areas inwhich officers had a better understanding of Promap. This suggests that further effortshould be expended to ensure that officers have a full appreciation of the approachbefore implementation begins.• There was a marked reduction of domestic burglary during the pilot study. This was amixed blessing. On the plus side, burglary declined more in the pilot area than in thecomparison area, and declined most during the shift for which the police had thegreatest opportunity to use, and reported most frequently using, Promap. On thenegative side, this meant that priorities other than domestic burglary came to the fore,with diminished use of the maps as a consequence. Further, the decline in burglarywas evident when implementation was at a rudimentary stage, and hence it is difficultto attribute the change to Promap. A fair test of the potential of prospective mappingmay only be realised when the method is tested across a range of areas and ideallywhen it is extended to cover a range of crime types. This was beyond the scope of thecurrent project, for which the aim was to determine the potential utility of the system.This research design is consistent with clinical trials of new pharmaceuticals, wherebydifferent phases of the trial are used to evaluate the efficacy of the drug. The initialphase, analogous to the approach adopted here, is essentially designed to uncover anyproblems with the drug and potential effectiveness rather than to demonstrate asystematic effect.An important element of the research was to gain feedback from those involved in the pilotingprocess to assess the potential of this type of system. A key message from this process wasthat the development of Promap is seen as a promising route towards intelligence-drivenpolice patrolling and the informed allocation of responsibilities within crime and disorderreduction partnerships. This is evidenced by the fact that (as noted above) officers enquiredabout the potential (immediate) development of the system for other types of crime, includingtheft from a vehicle.RecommendationsThe potential applications of Promap are manifold. It would have been too much to expectthat this potential could be fully exploited within a six-month trial period, where all parties werestarting from scratch. Listed below are a number of recommendations for implementation thatcould help realise improvements in operational practice, and a consequent reduction in crime.• Police officers located in diverse areas could be consulted in the development of themaps so that their operational usefulness can be tailored to different contexts. Mapscould distinguish between areas for which risks are increasing and those for whichthe level of risk is stable or declining. Which type of map is most useful may dependupon resources available and problem profiles. Bespoke mapping systems, locallyoptimised and responsive to tasking and co-ordination wishes are feasible.• All those tasked with acting upon the maps might be provided with regularinformation, which could become an integral part of the daily routine. If possible, newmaps should be provided two to three times per week and display shift-specific risks.viii

• To maintain momentum the system could either become part of the long-termroutines of operational policing or used for time-limited highly resourced operations. Amiddle ground, where attention is only paid when burglary is a priority, would limit theutility and understanding of the system.• Risks identified by Promap could usefully be combined with other forms of intelligenceto optimise operations. For example, high risk areas could be scanned for trends indata on modus operandi or for information on known offenders which could help focuspolice tactics. This type of analysis could, to some extent, be automated, therebyfreeing analyst time to allow a more thorough interrogation of patterns, and for themto consider the range of crime reductive responses possible.• It would be desirable to develop procedures that enabled evaluation of the systemwithout any demand on police time or resources. For example, an index of changes inthe patterns of crime clustering over time could be automatically produced andrecorded. Evaluation would also be facilitated by systematic documentation of theactivities and arrests made by beat officers, although any paperwork burden wouldneed to be minimised, perhaps through the use of a simple computerised datacollection tool.• In any mapping system, the risks identified are relative to the level of crime in aparticular area, and to the time frame selected for analysis. Hence it is possible toproduce maps showing areas of relatively high risk even when there is a lull in theunderlying problem. Thus, for the crime type(s) predicted some indication of theanticipated scale of the problem would be a useful additional feature of the system.• The authors believe that Promap should be extended to other crime types, such asvehicle crime and violence. Predictions may be produced for all crime (if subsequentempirical research suggests this to be sensible) and weighted by seriousness of eachcrime type. Because different types of crime might require different solutions, anindication of the anticipated volume of each type could be provided as an indication ofwhat to prioritise and when.As noted above, the use of forecasting systems such as Promap should encourage thecontinual consideration and reconsideration of operational tactics. Favoured approaches mayrequire rethinking when it is possible to more accurately predict where resources are mostneeded.The future?It is believed that the development of the techniques described will offer a step change in thepower and applicability of crime mapping as a tool of crime reduction. In perhaps fifteenyears, using techniques such as those discussed here and those as yet unrealised, predictivemapping could be available for all crime types; real time information on risk would therebybecome available to police patrols, where the seriousness of different crime types is weightedautomatically so that an optimal patrolling pattern is provided to each police vehicle tomaximise the total seriousness of crimes to be preventively patrolled. Using Promap typesystems in concert with Lab-on-a-chip forensic testing, where DNA and other tests would bepossible in police vehicles, would facilitate swift forensic identification of perpetrators ofcrimes not prevented, and patrolling informed by Promap would mean faster response timesto arrive before crime scenes are compromised for forensic purposes. In parallel withoptimised patrolling, Promap would deliver information about longer-term patterns andstabilities in crime and disorder to Crime and Disorder Reduction Partnerships, enabling themto put in place design and maintenance changes. Nothing in such a future is unfeasible evenwith today’s technology. It does however require an effort of imagination to discern thecentrality of prospective mapping to such a future. While the East Midlands study reportedhere was in most respects successful, the big prizes of intelligent crime reductive practice willbe won only through an integrated developed programme rather than a succession of ad hocpiecemeal projects.ix

1. IntroductionThis report represents a summary of a project concerned with short-term burglary prediction.The work was completed in three stages, a short initial research phase to develop andestablish the accuracy of the method, a short development phase during which the systemwas refined for use in one police Basic Command Unit (BCU) with tactical implicationsidentified and discussed, and finally a six-month field trial to test whether the system could beused in an operational context and how the police engaged with it.The structure of this report largely follows the chronology of the research, thereby providingthe reader with an impression of substance and sequence. The first section is a review of theliterature that informed the project, and the second the empirical research, initial developmentand testing of the system. The third section contains a discussion of the types of tacticaloption that were originally identified as having the potential to be used with the system, or atailored variant of it, and the wider ideology of the approach. The fourth section discussesfinal refinements to the system used in the field trial, and then reviews how the system waseventually used and police officers’ perceptions of its utility. Lessons learned regardingimplementation are also highlighted and discussed. In the penultimate section, changes inpatterns of crime coincident with the work are explored, and in the final section futuredirections and recommendations are discussed.Background and past researchCriminological research has demonstrated that crime is concentrated. For all crime typesanalysed, a small number of victims are repeatedly victimised and hence experience a largeproportion of crime (for reviews, see Pease, 1998; Farrell, 2005); a large proportion of crimeoccurs in a small number of areas; and a small number of offenders commit a large proportionof crime (e.g. Spelman, 1994). In relation to the geographical distribution of crime, thismanifests itself as spatial clustering, with ‘hotspots’ of crime such as burglary being a typicalcharacteristic of deprived areas (e.g. Johnson et al., 1997). These findings conform to what ismore generally known as the 80:20 rule. This pattern is not confined to crime but is a moregeneral phenomenon. For instance, a small proportion of the earth’s surface holds themajority of life on the planet, and a small proportion of earthquakes account for the majority ofdamage caused by them (Clarke and Eck, 2003).For burglary, the focus of the current research, the relationship between different types ofconcentration has also been studied. Specifically, are incidents of repeat burglaryvictimisation the work of a common offender, or do different offenders simply exploit the sameopportunities for crime? These explanations have been referred to within the literature as theboost and flag hypotheses, respectively (Pease, 1998). A number of approaches toinvestigating these hypotheses exist, but perhaps the most direct is to examine data fordetected offences. In their analysis of a sample of data for offenders detected for burglaryoffences, Everson and Pease (2001) demonstrate that 86 per cent of the incidents of repeatvictimisation were committed by the same offenders (see also Everson, 2003). Furthercorroborative evidence comes from studies in which offenders have been interviewedregarding their offending behaviour. Typical findings illustrate that around one in threeburglars admit to returning to the same property to commit a further offence (Gill andMathews, 1994; Ashton et al. 1998) and their reasons for so doing include the following:“the house was associated with low risk …., they were familiar with the features of thehouse …., to get things left behind or replaced goods.”Ericsson, 1995Perhaps the most succinct account was given by a Scottish burglar to Mandy Shaw. Uponasked why he returned, he replied “Big house, small van”. Thus, whilst recognising that somerepeat offences may be committed by unrelated offenders, a consensus of opinion isemerging that repeat victimisation is largely the work of the same offenders. A further findingthat supports this conclusion and which has immediate crime prevention implications is the1

time course of repeat victimisation (RV). Research consistently demonstrates that when RVoccurs it does so swiftly offering a limited but precise window of opportunity for intervention(e.g. Polvi et al., 1991). Risk is unstable. Thus, repeat victimisation may be said to be aspecial case of space-time clustering, events tending to occur swiftly at the same locations.Inspired by the precepts of optimal foraging theory, the authors have recently examinedwhether RV is part of a more general foraging pattern (Johnson and Bowers, 2004a). Thetheory, borrowed from behavioural ecology, is that when searching for resources, offenderswill aim to limit the time spent searching for suitable targets, whilst simultaneously seeking tomaximise the rewards acquired thereby minimising the associated risks. RV is arguably anexample of optimal foraging. A conjecture from Farrell et al. (1995) illustrates this. Farrell etal. suggest that:“a burglar walking down a street where he has never burgled before sees two kinds ofhouse – those presumed suitable and those presumed unsuitable. (The latteridentified by dint of an alarm, by occupancy, the presence of a barking dog, and soon). He burgles one of the houses he presumes suitable, and he is successful. Nexttime he walks down the street, he sees three kinds of house – the presumedunsuitable, the presumed suitable, and the known suitable. It would involve the leasteffort to burgle the house known to be suitable.”Farrell et al. (1995)Thus, offenders target those properties with which they are most familiar, and which combinegood rewards and acceptable risks. A natural extension of this strategy would be to target notonly those previously burgled and known to be suitable but also those houses that are mostsimilar to them, in terms of the likely risks and rewards and the effort involved in burglingthem. The first law of geography states that things which are closest to each other in spaceare the most similar. It follows that homes nearest to burgled houses may represent the nextbesttargets. For this reason, using data for the county of Merseyside and methodsdeveloped in the field of epidemiology (Knox, 1964), the authors conducted a series of studiesto examine whether the risk of burglary clusters in space and time more generally. That is,does the risk of burglary appear to be communicated from one property to another in muchthe same way as the behaviour of a disease? 1A series of confirmatory findings followed. In particular, for the area studied, the researchdemonstrated that the risk of burglary was communicated over a distance of about 400m andthis elevated risk endured for around one month (Johnson and Bowers, 2004a), after which itappeared to move to other nearby areas (Johnson and Bowers, 2004b). Additionally, adisproportionate increase in risk for those on the same side of the street as the burgled homewas evident. The communicability of risk varied by area, with risk appearing to be mostcommunicable in the most affluent of areas (Bowers and Johnson, 2005a), though somedegree of communicability was well-nigh universal.The practical implications of this programme of research are clear: crime reductive actionshould be directed towards the burgled home, and also to those nearby. However, oneconcern raised was the practicability of implementing such a strategy on a large scale.Consider that the implementation of a strategy for which every burgled household andneighbours within 400m received crime reduction attention would require substantialresources if implemented across an area such as a police Basic Command Unit. For obviousreasons, such a strategy is unlikely to generate much enthusiasm.What is required is a more precise method of generating reliably accurate predictions ofwhere crime will most likely next occur. Such a method should enable the efficientdeployment of resources. The rough location of a high concentration of crime could easily bepredicted by simply identifying a large urban area, but this would be of little operational value.The challenge, then, is to identify where a high concentration of crime will occur for arelatively small area. Considering the findings in relation to crime concentration, for an1 The authors do not suggest that burglary exudes a bacillus but that the clustering of events in space and time mightsuggest that it does so.2

optimally calibrated system, the goal might be to be able to identify for each day or police shiftthe 20 per cent of locations in which 80 per cent of crime was most likely to occur. The extentto which this can be achieved will, of course, be affected by the degree to which crimeconcentrates in spatial and temporal terms, and the degree to which spatio-temporalsequences can be captured as maps.The key conceptual point in the enterprise is that past events are not themselves mapped inthe conventional way, but contribute risks to locations over time and space to diminishingextents. This approach has come to be known as prospective mapping (see Bowers et al.,2004; Johnson et al., 2005). The system uses recent historic crime data to generateforecasts which can be displayed using a Geographical Information System (GIS) andoverlain on an Ordinance Survey (OS) map of the relevant policing area, allowing crimereductive resources to be deployed to those areas when and where they will be most needed.The model is calibrated according to the dimensions of the communicability of risk for thearea considered. Early evaluation of the system has shown that for the county of Merseysidethe locations of a large percentage of burglaries (64%-80%) (occurring up to three days afterthe generation of the forecasts) can be identified within a small fraction of the total areaconsidered (20%). This level of accuracy outperformed existing methods of hotspot mapping(Bowers et al., 2004), which themselves outperformed predictions based on police officersperceptions of high risk areas (McLaughlin et al., forthcoming), and the efficiency of aprediction slowly diminished after three days (Johnson et al., 2005) as was anticipated.Aims and objectivesThe results of the above studies show considerable promise for crime reduction. However,the external validity of the findings, that is to say how generalisable they are to other areasand different contexts, remained unknown. The main aims of the current project were sixfold.1. To see if the patterns of communicability discussed above are observed in anotherarea of the UK, namely the East Midlands.2. To see how this pattern varies between areas within the East Midlands.3. To investigate how the police and their crime reduction partners currently deploycrime reduction resources, what they do and how they identify where to implementstrategies, as a baseline against which to consider the action implications ofpredictive mapping.4. Given that crime does cluster in space and time as observed on Merseyside, toevaluate the predictive accuracy of prospective mapping in these areas by comparingit with the systems currently used there, and against what would be expected ifpatrols were directed to areas randomly5. To develop and field-test the system in one police BCU to explore its feasibility in anoperational setting on a routine basis, and to identify obstacles to implementation.6. To evaluate the impact on crime of crime reduction strategies informed by the systemin the area.In the chapters that follow, the authors will discuss the results of the research and illustratehow the theory of prospective mapping (hereafter, Promap) evolved into a tactical entity thatwas used in an operational context.3

2. Testing the generalisability of prospectivemappingAs discussed above, research concerned with geographical hotspots of crime demonstratesthat over a given period of time crime is spatially concentrated (see, for example, Eck andWeisburd 1995). Other research (e.g. Farrell and Pease, 1994) demonstrates that within aspecific geographical region, crime also exhibits temporal clustering, with incidence ratesincreasing with the onset of the winter, and diminishing around the spring. However, unlessthe risk of victimisation is communicable, one would expect these spatial and temporalclusters to be independent of each other, with the incidence of new series of crime eventsfailing to exhibit localised spatial and temporal increases. Simply put, the communicability ofburglary risk may be demonstrated only by showing that burglary clusters in both space andtime. Conformity to this pattern would be evident if houses near the burgled home werevictimised shortly afterwards more than would be expected on the basis of chance. Thosereaders not concerned with technical details on the demonstration of clustering should skip tothe beginning of the next chapter at this point.Empirical research concerned with the space-time clustering of events was first conducted byKnox (1964) to study epidemics of leukaemia. The rationale underlying the Knox test is todetermine whether there are more events that occur close in space and time than would beexpected on the basis of a random distribution. To do this, each event is compared with everyother and the distance and time between them recorded. For n cases, this generates ½n (n-1) pairings. A contingency table, such as shown in Table 2.1, with i columns and j rows isthen populated. The spatial and temporal increments (or bandwidths) used in the rows andcolumns can be arbitrary, although they should be so defined as to allow specific hypothesesinformed by the underlying theory to be tested. For instance, in the case of crime the questionconcerns the distance over which crime has an impact and for how long? The bandwidthsselected should have relevance to operational policing.Table 2.1: Knox contingency table example1 month 2 months 3 months …………..100m n 11 n 21 n 31200m n 12 n 22 N 32300m n 13 n 23 N 33400mn 14 n 24 n 34……..……..Note: The mathematical notation n ij refers to the observed frequency for the cell occurring in column i and row j.Thus, n 11 refers to the cell in column 1, row 1.The contingency table generated can be compared with a chance distribution. Onecomplication is that the assumption of independence of observations, a criterion for mostinferential statistical methods, is violated. However, Knox suggested that in the absence of aspace-time interaction, the statistical distribution of the expected values for the cells of thecontingency table would conform to a Poisson distribution and could be computed in thesame way as a Chi-Square test, using the marginal totals of the table. Thus,4

e ij = n .j x n i.nWhere, e ij is the expected value for each cell, n ij is the observed, and n i and nj are the row andcolumn totals respectivelyThe results of the analysis can be interpreted by inspecting adjusted residuals, computed foreach cell using the following formula (see, Agresti, 1996, pp31-32):r ij =n ij – e ij[e ij (1-row proportion of n i+ )(1-column proportion of n +j )] 1/2Thus, the adjusted residuals are a measure of the difference between the observed andexpected values for each cell. The adjusted residuals have a mean value of zero and astandard deviation of one, hence adjusted residual scores exceeding values of two (or evenmore stringently three) are considered statistically significant. The null hypothesis is thatevents are not clustered in space and time. Negative values indicated that there are fewerevents occurring within a particular space-time interval.The majority of research concerned with crime (Johnson and Bowers, 2004; Townsley et al.,2003) has used the Knox approximation to examine space-time clustering. However, analternative approach, for which the independence of observations is not a requirement, mayalso be computed (Johnson et al., 2006). This approach uses Monte-Carlo simulation togenerate an expected distribution, rather than using the marginal totals (Besag and Diggle,1977). To do this, the data are permuted, in effect mixing up the dates and locations acrossthe events. Thus, the dates are randomly shuffled using a pseudo-random number generator,whilst the spatial locations remain fixed. The hypothesis is that if there is statisticallysignificant space-time clustering in the data then there should be more events observed tooccur close in space and time than for 95 per cent of the random permutations generated.The process of generating permutations is repeated (iterated) a number of times, typicallyaround 999, a new contingency table generated each time and compared with thecontingency table for the observed distribution. The probability that the observed value foreach cell occurred on the basis of chance may be calculated using the following formula (seeNorth, 2002):n − rank + 1p =n + 1Where n is the number of simulations, and rank is the position of the observed value in a rankordered array for that cellOne possible reservation about the p-values generated using the Monte-Carlo approach (andthe Knox residuals) is that they are not as readily interpretable as one might like.Interpretation of the results can be aided by deriving a simple measure of effect size. 2 In thiscase, a Knox ratio, which contrasts the observed value for each cell with the average‘simulated’ value for that cell (or the expected value derived using the marginal totals of thetable), was computed. Thus, a Knox ratio of 2.0 so derived would indicate that twice as manyburglaries occurred within a particular distance and time of each other than would on averagebe expected on the basis of a chance distribution. A value of one would indicate that theresult conformed to what would be expected on the basis of chance, and values of less thanone that fewer burglaries occurred within a particular distance and time of each other thanone would expect. Which measure of central tendency is best used to compute the odds ratiodepends upon the distribution of the simulated data. If the data are skewed, or there are2 The Knox residuals can be interpreted in the same way to some extent but are not as simple to understand. Insimple terms, the larger the Knox residual the larger the effect.5

extreme values, the mean can misrepresent the distribution. For this reason, the authors useboth the mean and median values for each cell. On the whole, for the data analysed the twotypically converge and hence for the results that follow, the authors will report only resultsderived using the median values.This approach is to be preferred over the Knox approximation, particularly as no assumptionsregarding the statistical distribution are required. However, limitations in computing powerhave meant that where a large number of events are to be analysed, the process can takesome time. This is particularly true where the dimensions of the contingency table or thenumber of events analysed are large. Fortunately, computing power is now sufficient forMonte-Carlo simulation of this kind to be completed for small data sets (e.g. n=1000) and a 10x 10 contingency table fairly rapidly, and for larger data sets (e.g. n=2000) over an hour or so.Larger contingency tables or data sets can still take a matter of days, but are now easilycomputed by those prepared to be as patient as the present writers!Is the risk of burglary communicable in the East Midlands?In the sections that follow, the results for each policing area that was analysed in the EastMidlands, namely Mansfield and Ashfield Sectors in Nottinghamshire, Alfreton (‘A’) Division inDerbyshire, and Corby and Wellingborough Sectors in Northamptonshire, will be presented.To allow easy direct comparisons the results will be presented here in tabular form. Anumber of analyses were conducted for each data set using both the Knox and Monte-Carlomethods described in the previous section. For instance, analyses have been conductedusing months as the temporal bandwidth whereas others have used weeks. For each areamonthly analyses will be presented for both Knox and Monte-Carlo approaches. As willbecome apparent these varied little. Thus, analyses for the Knox approximation only will bepresented looking at the communication of risk over weekly intervals. In all analyses, for thesake of simplicity, repeat victimisation is excluded. Thus, the results consider the distanceover which the risk of a household’s burglary is communicated to different households and forhow long this endures. Because of the exclusion of repeats against the same household, thebenefits of patrolling and resourcing patterns based upon propinquity in time and space to aburgled home will be considerably understated in what follows.To anticipate the results, it would appear that as a general rule the risk of burglary iscommunicable up to a distance of around 400m for at least one month. There was no area forwhich this was not the case, hence universal application of the 400m one-month rule is in nocase inappropriate, it merely excludes a proportion of other high-risk homes. The sensitivityanalyses conducted to examine the duration of the elevation in risk suggested that thisextended beyond one month in most cases, up to around eight weeks. This will subsequentlybe referred to as the classic profile. Those not wishing to understand the detail of areadifferences should skip to the next section, in which the value of prospective mapping inoperational context is explored.Mansfield Police SectorFor Mansfield, the data analysed cover the period January-December 2004 and comprised1,156 burglary events. The results of the monthly Knox and Monte-Carlo analyses are shownas Tables 2.2 and 3.2 respectively. The results are consistent in showing that the risk ofburglary communicates up to a distance of 200m and endures for one month. Additionally,the Monte-Carlo results suggest that for houses within 100m of burgled homes the riskendures for up to two months. The same (two-month) result is apparent for the Knox residualanalysis, although it marginally fails to be statistically reliable.6

Table 2.2: Knox ratios for Mansfield (values in bold are statistically significantaccording to the residual scores (not shown), N=1,566)MONTHS1 2 3 4 5 6100 1.26 1.10 0.97 0.85 1.02 1.03200 1.08 0.99 0.93 1.04 0.92 1.01300 1.00 1.07 1.01 1.01 1.01 1.13400 1.03 0.96 1.07 1.04 1.01 1.06500 1.01 1.06 0.97 0.98 1.11 1.05600 1.03 1.02 0.98 1.08 0.98 1.03700 1.03 1.04 0.96 1.00 1.01 1.02800 1.04 1.01 0.97 0.98 1.00 1.00900 1.03 1.06 0.97 0.96 0.98 1.031000 1.06 1.01 0.96 1.02 0.99 1.02A number of other cells are statistically significant in each table, namely those for 1,000m andone-month and 500m and five months. Due to the results for a large number of cells beinganalysed, some cells (60*0.05=3) will have significant values by chance. For this reason theywill be discussed no further. In contrast, the analyses for the cells in the top left of the tablesare based on an a priori hypothesis and are of clear relevance.Visual inspection of the Knox ratios in Table 2.3, shows that around 25 per cent moreburglaries occur within 100m and one month of burgled homes than would be expected onthe basis of chance, eight per cent more between 101-200m. 12 per cent more up to 100mand between 30-60 days. Otherwise, the number of burglaries approximate what would beexpected on the basis of chance. (Note that the Knox residuals cannot be interpreted aspercentage deviations from chance as they are actually z-scores, see preceding section).Table 2.3: Monte-Carlo results for Mansfield (values in bold are statistically significant,N=1566)MONTHS1 2 3 4 5 6100 1.26 1.12 0.92 0.90 0.99 1.08200 1.08 0.99 0.93 1.02 0.95 0.99300 1.00 1.06 1.00 1.01 1.00 1.17400 1.03 0.98 1.08 1.03 1.00 1.03500 1.01 1.05 0.96 1.00 1.10 1.06600 1.03 1.02 0.97 1.09 0.97 1.03700 1.03 1.04 0.94 1.02 1.00 1.01800 1.04 1.01 0.99 0.96 1.01 0.98900 1.03 1.05 0.96 0.98 0.98 1.001000 1.06 1.00 0.97 1.01 1.00 1.03To increase the sensitivity of the results, an additional Knox analysis was conducted toexamine the temporal pattern in more detail. Thus, weekly rather than monthly patterns wereanalysed. As noted above, these were computed using the Knox approximation rather thanMonte-Carlo simulation.7

Table 2.4: Weekly Knox ratios for Mansfield (values in bold are statistically significantaccording to the residual scores (not shown), N=1566)WEEKS1 2 3 4 5 6 7 8 9 10 11 12100 1.43 1.29 1.31 1.23 1.21 1.20 1.17 1.14 1.13 1.06 1.03 0.98200 1.38 1.20 1.14 1.09 0.99 0.94 0.94 0.95 0.98 1.01 0.99 0.96300 1.09 1.04 1.00 1.00 1.01 1.04 1.06 1.09 1.05 1.03 1.05 1.00400 1.07 1.06 1.02 1.04 1.00 0.99 0.98 0.94 0.97 1.01 1.06 1.05500 0.97 0.98 1.00 1.01 1.01 1.05 1.05 1.08 1.06 1.03 1.00 0.95600 1.03 1.10 1.06 1.04 1.03 0.99 0.99 1.00 1.03 1.01 1.02 1.01700 0.99 1.01 0.99 1.02 1.03 1.04 1.04 1.03 1.05 1.04 1.03 1.00800 1.04 1.02 1.03 1.05 1.04 1.03 1.03 1.00 1.02 1.01 1.00 0.98900 1.02 1.02 1.06 1.04 1.04 1.05 1.02 1.04 1.05 1.03 1.03 1.001000 1.10 1.09 1.07 1.07 1.04 1.00 1.00 0.98 1.02 1.00 0.99 0.99The results, shown as Table 2.4, suggest that the risks for houses within 100m of a burgledhome endure for up to nine weeks, and for those within 200m four weeks. Thus, the resultsof the monthly and weekly analysis reveal essentially the same pattern.Wellingborough Police SectorFor Wellingborough, the data cover the period for which the most recent data were availableat the time of analysis, in this case January-December 2003. There were 1,350 burglariesover this period. Tables 2.5 and 2.6 show the Knox and Monte-Carlo analyses. A similarpattern emerges. The results suggest an elevated risk to those proximate to the burgled homefor around one month.Table 2.5: Knox ratios for Wellingborough (values in bold are statistically significantaccording to the residual scores (not shown), N=1350)MONTHS1 2 3 4 5 6100 1.63 1.04 0.70 0.82 0.96 0.64200 1.37 1.00 0.85 0.78 0.89 0.85300 1.27 0.96 0.90 0.82 0.89 0.93400 1.20 0.98 0.96 0.94 0.93 0.89500 1.17 1.03 0.94 1.01 0.92 0.93600 1.10 1.04 1.00 0.92 0.93 0.87700 1.01 1.05 0.99 0.96 0.94 0.93800 1.08 0.98 1.03 1.02 0.98 0.91900 1.01 0.97 0.95 0.95 1.02 0.951000 0.99 1.00 1.04 0.98 0.93 1.05The results of the Monte Carlo analysis show that 57 per cent more burglaries occurred within100m and one month of each other than would be expected on the basis of chance. A largenumber of burglaries also occur up to 400m and one month of each other, thereafter the Knoxratios diminish although they remain statistically significant up to 700m.To examine the pattern in more detail, a weekly analysis was conducted. The results, shownas Table 2.7, suggest a similar pattern but that the elevated risks endure for more than onemonth.8

Table 2.6: Monte-Carlo results for Wellingborough (values in bold are statisticallysignificant, N=1350)MONTHS1 2 3 4 5 6100 1.57 0.97 0.77 0.82 1.00 0.72200 1.35 1.01 0.85 0.77 0.91 0.86300 1.26 0.96 0.90 0.81 0.90 0.93400 1.21 0.99 0.96 0.91 0.91 0.92500 1.16 1.03 0.93 1.00 0.94 0.95600 1.10 1.03 1.00 0.93 0.92 0.86700 1.01 1.05 0.98 0.98 0.95 0.91800 1.08 0.98 1.03 1.02 0.99 0.89900 1.01 0.98 0.95 0.96 0.99 0.941000 0.98 1.01 1.04 0.98 0.92 1.05These findings illustrate a problem with analyses where a certain level of aggregation isemployed. This is known as Simpson’s paradox (Simpson, 1969). This occurs where dataaggregated up to a fairly large unit of analysis such as one month, mask subtle trendsapparent where data disaggregated to a finer level of resolution (such as one week) are used.In this case, it would appear from the analysis shown in Table 2.7 that the risk to houses up to600m from victimised homes endures for around six to seven weeks rather than four. In anyevent, it is apparent that the risks are consistently greater within the first month.Table 2.7: Weekly Knox ratios for Wellingborough (values in bold are statisticallysignificant according to the residual scores (not shown), N=1350)WEEKS1 2 3 4 5 6 7 8 9 10 11 12100 1.82 1.78 1.76 1.67 1.53 1.39 1.22 1.06 0.97 0.87 0.78 0.76200 1.64 1.56 1.46 1.39 1.27 1.18 1.09 1.03 1.00 0.96 0.89 0.86300 1.52 1.43 1.37 1.29 1.20 1.12 1.04 1.00 0.94 0.91 0.88 0.88400 1.39 1.28 1.23 1.21 1.15 1.09 1.04 0.99 0.95 0.94 0.96 0.95500 1.27 1.23 1.19 1.18 1.13 1.11 1.06 1.04 1.01 0.98 0.97 0.95600 1.11 1.08 1.09 1.11 1.08 1.08 1.06 1.05 1.03 1.03 1.02 1.01700 1.02 1.01 1.02 1.00 1.02 1.04 1.06 1.06 1.04 1.01 1.00 1.00800 1.16 1.08 1.10 1.07 1.05 1.04 1.02 1.00 0.99 1.00 1.00 1.03900 1.04 1.03 1.01 1.01 1.01 0.99 0.98 0.97 0.98 0.98 0.98 0.981000 1.00 1.01 1.01 0.99 0.97 0.97 0.96 1.00 1.01 1.01 1.04 1.04Ashfield Police SectorFor Ashfield, the data cover the period January-December 2004 and a total of 1,012 burglaryevents. The results of the monthly Knox and Monte-Carlo analyses are shown as Tables 2.8and 2.9 respectively.9

Table 2.8: Knox ratios for Ashfield (values in bold are statistically significantaccording to the residual scores (not shown), N=1012)MONTHS1 2 3 4 5 6100 1.19 0.96 1.04 1.04 0.96 0.79200 1.17 0.99 0.94 0.97 1.00 0.91300 1.18 0.96 0.99 0.99 0.95 1.00400 1.11 0.96 0.96 1.11 0.94 1.01500 1.02 1.03 0.99 1.05 0.96 1.00600 1.09 1.04 1.01 1.07 0.97 0.98700 1.12 0.96 1.07 1.05 0.91 0.89800 0.99 1.01 1.02 0.98 0.96 1.06900 0.96 0.96 1.08 0.96 1.05 0.981000 0.97 1.02 1.02 1.01 1.06 0.92The results are again very consistent. Both analyses suggest that the risk of burglarycommunicates over a distance of up to 700m and endures for one month. Interestingly, therisk communicated to houses within 401-500m appears to be non-significant. This is difficultto explain. It could be an expression of the spatial distribution of targets across the area, orperhaps suggests the existence of a natural boundary that generates this pattern bydiscouraging some offenders from travelling from one area to a second nearby. Moreresearch would be required to understand this effect. Nevertheless, the results clearlydemonstrate that the risk of burglary is communicable.Table 2.9: Monte-Carlo results for Ashfield (values in bold are statistically significant,N=1012)MONTHS1 2 3 4 5 6100 1.18 0.96 1.03 1.05 0.96 0.80200 1.17 0.98 0.95 0.97 1.00 0.92300 1.17 0.97 0.98 0.99 0.94 1.01400 1.11 0.95 0.97 1.11 0.95 1.00500 1.02 1.03 0.98 1.05 0.96 0.98600 1.08 1.04 1.01 1.07 0.98 0.97700 1.12 0.96 1.07 1.06 0.90 0.90800 0.99 1.00 1.03 0.97 0.96 1.06900 0.96 0.96 1.08 0.94 1.07 0.981000 0.97 1.01 1.03 1.02 1.07 0.92As before, an additional Knox analysis was conducted to examine the temporal trend in moredetail. The results, shown as Table 2.10, suggest a similar pattern of risk, but additionallysuggest that houses between 401-500m of burgled homes are at an elevated (albeitmarginally non-significant) risk of victimisation for up to two weeks after an initial burglary.Thus, this suggests that the risk of victimisation is, in general, communicable up to 700m, butfor slightly different time periods.10

Table 2.10: Weekly Knox ratios for Ashfield (values in bold are statistically significantaccording to the residual scores (not shown), N=1012)WEEKS1 2 3 4 5 6 7 8 9 10 11 12100 1.61 1.44 1.28 1.22 1.06 1.02 0.96 0.96 0.96 0.97 0.97 0.97200 1.46 1.29 1.25 1.18 1.08 1.00 0.99 0.99 0.98 1.01 0.94 0.93300 1.43 1.39 1.23 1.17 1.08 1.05 0.99 0.98 0.94 0.97 0.95 1.00400 1.20 1.16 1.15 1.12 1.08 1.06 1.00 0.96 0.96 0.95 0.95 0.98500 1.13 1.10 1.05 1.03 1.00 0.99 0.97 1.01 1.02 0.96 0.97 0.96600 1.20 1.16 1.13 1.09 1.07 1.05 1.01 1.04 1.04 1.04 1.04 1.05700 1.35 1.20 1.16 1.13 1.06 1.02 1.01 0.96 0.96 0.99 1.01 1.04800 1.06 1.03 1.04 1.00 0.98 0.96 0.97 0.97 0.98 1.01 1.02 1.05900 0.96 0.98 0.97 0.97 0.96 0.91 0.90 0.94 0.96 1.00 1.08 1.121000 0.93 0.96 0.97 0.96 0.99 0.99 1.00 1.03 1.03 1.06 1.03 1.06Corby Police SectorFor Corby, the most recent data available at the time of analysis covered the period January-December 2003. There were only 429 burglaries for this period in Corby. The results, shownas Tables 2.11 and 2.12, are again very consistent. Unlike the above analyses, while moreburglaries occur within 100m and one month of each other than would be expected on thebasis of chance, the difference is not statistically reliable. However, this may be due to thelow sample size (e.g. the expected cell count for that cell was five times lower than the samecell for Mansfield), particularly for that cell, rather than reflecting an irregularity in the patternof the communicability of risk. As with the analyses for Mansfield the analyses suggest thatthe risk of burglary communicates over a distance of up to 700m and endures for one month,although houses within 401-500m appear not to be at significantly heightened risk.Table 2.11: Knox ratios for Corby (values in bold are statistically significant accordingto the residual scores (not shown), N=429)MONTHS1 2 3 4 5 6100 1.17 1.07 1.11 1.02 1.06 0.87200 1.20 0.99 1.22 1.02 0.89 0.96300 1.13 0.94 0.90 1.12 0.96 0.92400 1.22 1.01 1.02 1.05 1.10 0.96500 1.04 1.04 1.02 1.09 1.09 1.03600 1.11 1.08 1.11 1.02 1.02 1.02700 1.12 1.01 1.11 0.91 0.98 1.03800 1.02 1.06 1.07 0.98 1.13 1.06900 1.02 1.04 1.05 1.03 0.93 1.111000 1.10 0.91 1.09 1.03 1.06 1.06As with the results for Mansfield, the Monte-Carlo analysis suggests that around 20 per centmore burglaries occurred within 101-200m and one month of a burgled home. Alsoconsistent with the findings for Mansfield, the risk to houses within 401-500m of burgledproperties appears to be elevated yet non-significant.For completeness, an additional Knox analysis was conducted to examine the temporal trendin more detail. However, as this analysis generated a larger number of cells than the monthlyanalysis and there was such a low volume of crime for this area, the results were consideredunreliable and will not be discussed further. Because of the essential similarity of the data11

across areas it is thought likely that the same patterns underlie the Corby data, andapplication of predictive mapping in low crime areas is by no means regarded as unprofitable.Table 2.12: Monte-Carlo results for Corby (values in bold are statistically significant,N=429)MONTHS1 2 3 4 5 6100 1.18 1.02 1.12 1.02 1.09 0.87200 1.20 0.96 1.23 1.03 0.88 0.96300 1.13 0.93 0.86 1.16 0.91 0.96400 1.22 1.03 1.04 1.07 1.09 0.96500 1.04 1.04 1.04 1.07 1.05 1.10600 1.11 1.12 1.08 1.01 1.04 1.00700 1.12 1.03 1.07 0.93 0.99 1.03800 1.02 1.07 1.06 1.02 1.14 1.07900 1.03 1.07 1.08 1.02 0.97 1.101000 1.09 0.92 1.09 1.04 1.06 1.06Alfreton ‘A’ DivisionFor ‘A’ Division in Derbyshire, the period for which the most recent data were available at thetime of analysis was April 2003-March 2004. There were 1,703 burglaries during this period.More recent data are now available but were not used for the analyses in this section of thereport. Tables 2.13 and 2.14 show the Knox and Monte-Carlo analyses. A similar patternemerges to those presented above, although the risk appears to communicate over a greaterdistance than for the other areas. Thus, the results suggest an elevated risk to those up to900m from burgled homes for around one month.Table 2.13: Knox ratios for ‘A’ Division (values in bold are statistically significantaccording to the residual scores (not shown), N=1703)MONTHS1 2 3 4 5 6100 1.37 0.93 0.91 0.93 0.85 0.76200 1.29 0.93 0.91 0.93 0.93 0.96300 1.19 1.00 0.94 0.94 0.83 0.90400 1.17 0.97 0.99 0.94 0.85 0.91500 1.08 0.95 0.91 0.98 0.85 1.02600 1.09 1.02 0.95 0.95 0.89 0.95700 1.07 1.00 0.95 0.95 0.88 1.07800 1.06 1.02 0.94 0.98 0.98 0.95900 1.16 1.03 0.97 0.91 0.93 0.891000 1.04 1.03 0.99 1.01 0.86 0.94As with the other analyses, the Knox ratios shown in Table 13.2 demonstrate a pattern ofdistance decay. That is, the risk of burglary is greatest nearest to burgled homes, after whichit remains elevated but clearly diminishes. Expressed in a slightly different way, just under 40per cent more burglaries occurred within 100m and one month of each other than would beexpected on the basis of chance. This is the highest concentration observed for that cellacross the data sets analysed. In contrast, around eight per cent more burglaries occurredbetween 400-800m and one month of each other than would be expected.12

Table 2.14: Monte-Carlo results for ‘A’ Division (values in bold are statisticallysignificant, N=1703)MONTHS1 2 3 4 5 6100 1.36 0.93 0.93 0.94 0.85 0.79200 1.27 0.95 0.91 0.93 0.93 0.93300 1.20 0.99 0.95 0.95 0.82 0.90400 1.18 0.96 0.98 0.93 0.85 0.95500 1.08 0.96 0.93 0.95 0.86 1.02600 1.08 1.01 0.97 0.95 0.88 0.99700 1.08 0.97 0.96 0.94 0.88 1.04800 1.06 1.01 0.95 0.98 0.98 0.96900 1.16 1.03 0.97 0.92 0.92 0.891000 1.04 1.05 0.96 1.02 0.86 0.92To examine the pattern in more detail, a weekly analysis was conducted. The results, shownas Table 2.15, suggest a similar pattern but that again the elevated risks endure for more thanone month in some cases (e.g. 101-200m), less in others (e.g. 601-700m).Table 2.15: Weekly Knox ratios for ‘A’ Division (values in bold are statisticallysignificant according to the residual scores (not shown), N=1703)WEEKS1 2 3 4 5 6 7 8 9 10 11 12100 2.38 1.77 1.54 1.40 1.08 0.97 0.96 0.96 0.96 0.90 0.92 0.85200 1.63 1.41 1.38 1.30 1.15 1.10 0.99 0.98 0.96 0.96 0.94 0.92300 1.38 1.35 1.24 1.21 1.11 1.05 1.02 0.99 0.97 0.97 0.99 0.96400 1.26 1.22 1.21 1.18 1.16 1.11 1.06 0.97 0.94 0.91 0.93 0.95500 1.05 1.10 1.08 1.07 1.04 1.03 1.00 0.97 0.96 0.94 0.93 0.91600 1.14 1.13 1.09 1.08 1.04 1.02 1.03 1.01 1.03 1.02 0.97 0.98700 1.18 1.09 1.06 1.06 1.04 1.06 1.05 1.00 0.98 0.94 0.93 0.95800 1.15 1.11 1.08 1.06 1.02 1.02 1.03 1.04 1.03 0.97 0.93 0.94900 1.21 1.20 1.17 1.16 1.12 1.10 1.10 1.09 1.05 1.00 0.99 0.951000 1.20 1.05 1.04 1.03 1.02 1.04 1.04 1.05 1.04 1.03 1.03 0.99Summary of the patterns so farThe above results demonstrate that for all areas analysed, there was clear evidence that therisk of burglary is communicable, although the distances over which this occurred varied byarea. In relation to the time over which this elevated risk endured, this tended to be at leastone month, but in some cases extended over longer intervals. One way of summarising thepatterns observed is presented as Table 2.16. Each cell of the table is shaded to indicate forhow many areas the particular cell had an over-representation of burglary pairs. This allows aquick comparison to see for how many areas the risk of burglary communicated up to say100m and one month, 201-300m and two months, and so on. It is clear that the mostcommon pattern was for the risk of burglary to extend up to 400m and for one month,although for some areas the risk communicated over greater distances, even up to 1,000m.Considering the unexplained results that were not anticipated by the theory (e.g. 200m andthree months for Corby), it is clear that there was no regularity in these results, suggestingthat they probably reflect spurious correlations or statistical artefacts.Each cell also contains the average Knox ratio for that cell, calculated across the five datasets. Not surprisingly the results correlate with the colour coding, with those cells shaded in13

the darkest colours having the highest values. It is also apparent that in general there is apattern of distance decay, such that the highest Knox ratios are for the shortest spatialintervals.Table 2.16: Summary or the analyses concerned with the communicability of risk (cellvalues are Knox ratios generated using Monte-Carlo simulation)MONTHS1 2 3 4 5 6100 1.26 1.00 1.01 0.99 0.97 0.87200 1.20 0.97 0.99 0.98 0.95 0.94300 1.17 0.98 0.95 1.02 0.92 1.01400 1.18 0.97 1.01 1.05 0.97 0.99500 1.04 1.02 0.98 1.02 0.99 1.03600 1.08 1.05 1.01 1.04 0.97 0.99700 1.08 0.99 1.02 1.00 0.93 0.98800 1.04 1.02 1.01 0.98 1.01 1.03900 1.03 1.01 1.03 0.96 1.00 0.991000 1.04 1.00 1.02 1.02 1.01 0.97Key 4-5 2-3 0-1Thus, it would appear that as a general rule the risk of burglary is communicable up to adistance of around 400m for at least one month. There was no area for which this was not thecase, hence universal application of a 400 metres one month rule is in no case inappropriate,merely that it excludes a proportion of other high-risk homes. The sensitivity analysesconducted to examine the duration of the elevation in risk suggested that this extendedbeyond one month in most cases, up to around eight weeks. This will subsequently bereferred to this as the classic profile.Spatial variation?Compared to the other areas considered, ‘A’ Division covers a large geographic area, thisbeing approximately 150 square miles. For this reason, additional analyses were conductedfor areas nested within the BCU. In so doing, the analyses control for another example ofSimpson’s paradox (Simpson, 1969), commonly referred to as the Modifiable Areal UnitProblem within the research literature (e.g. Openshaw, 1995). In this case, the problem isthat it is possible, even likely, that patterns evident at the aggregate level (i.e. across all fiveareas) differ from those that would be apparent at the local level. Thus, further analyses wereconducted to examine the patterns at the local level for five smaller areas within ‘A’ Division.To do this, analyses were conducted for each of the five policing sections within ‘A’ Division.However, one problem with conducting such analysis is that the geographical policingboundaries used are usually artificial and hence if the data are analysed for one area at atime using the existing boundaries, this can create what is known as an edge effect. Theproblem that arises is that patterns which occur across and around the boundary can remainundetected. Where boundaries are defined by natural features such as rivers or othertopological barriers this of course is not a problem.To reduce the likelihood of encountering this problem, a 1km buffer was generated aroundeach area and the data within the larger area analysed. In this way, any patterns of offendingat the edges of the policing boundaries should be detected. The five different policing areasand the buffer zones used are shown in Figure 2.1.14

Figure 2.1: The five policing areas in ‘A’ Division (right panel shows the buffer zoneused in the analyses)The results of the analysis for each area are shown in Tables 2.17-2.21. For area 1, it wouldappear that the risk of burglary communicates up to around 400m and particularly for oneweek after an initial event. This is clearly different to the ‘A’ Division wide analyses presentedabove. It is also apparent that many more burglaries occur within 100m and one week ofeach other than would be expected if there were no communication of burglary risk.Table 2.17: Weekly Knox analysis for area 1 (values in bold are statistically significantaccording to the residual scores (not shown), N=409)WEEKS1.00 2.00 3.00 4.00 5.00 6.00 7.00 8.00 9.00 10.00 11.00 12.00100 2.71 0.86 1.14 0.86 0.68 0.86 1.07 1.23 0.59 0.89 1.12 1.08200 1.65 1.22 0.91 0.89 1.00 1.23 1.16 1.03 1.05 0.94 1.31 1.00300 1.44 1.17 1.14 1.22 1.09 0.84 1.00 1.01 1.02 0.77 1.13 1.10400 1.32 1.16 1.04 1.15 1.21 1.00 0.89 1.05 0.89 0.79 1.09 0.94500 0.98 1.17 1.17 0.91 1.10 1.07 0.92 0.97 1.28 1.14 0.85 0.93600 1.22 1.00 0.92 0.99 1.02 1.18 0.79 1.04 1.00 1.02 1.05 1.08700 1.09 0.97 0.97 0.97 1.10 1.13 1.04 0.93 0.96 0.98 0.98 1.08800 1.17 1.10 0.99 1.08 1.06 1.07 1.24 1.07 0.82 0.83 0.90 0.96900 1.00 0.96 1.04 1.06 1.07 1.14 0.99 0.93 1.24 0.96 0.80 0.801000 0.94 1.01 1.00 1.17 0.92 0.91 1.09 0.91 1.15 1.07 0.90 0.90The results for area 2, shown in Table 2.18, reveal a similar pattern of results, although thepattern of distance decay is less dramatic than for area 1.15

Table 2.18: Weekly Knox analysis for area 2 (values in bold are statistically significantaccording to the residual scores (not shown), N=270)WEEKS1.00 2.00 3.00 4.00 5.00 6.00 7.00 8.00 9.00 10.00 11.00 12.00100 1.90 1.29 0.70 0.75 1.16 0.89 1.00 1.26 0.83 1.00 0.72 1.18200 1.49 1.26 1.15 0.91 0.77 0.66 1.14 1.05 0.95 0.95 0.71 0.88300 1.38 1.11 0.86 0.84 0.85 0.82 1.24 1.15 1.00 1.18 1.19 0.81400 1.22 1.18 1.09 0.88 1.06 1.06 1.04 0.96 0.92 0.80 0.79 1.16500 1.02 0.99 0.88 0.93 1.03 1.05 0.90 1.01 0.93 0.91 0.99 0.93600 1.31 0.96 1.00 1.09 0.89 0.90 0.88 0.92 0.79 0.99 0.79 1.08700 1.12 0.99 0.87 0.74 1.02 1.05 1.06 0.82 0.90 0.98 1.01 0.94800 1.12 1.15 0.81 1.04 1.02 1.02 0.99 1.15 1.01 0.95 0.94 1.09900 1.03 1.17 0.98 1.10 0.98 0.93 0.99 1.12 0.95 0.87 1.06 0.911000 1.11 0.96 1.01 0.88 1.15 1.11 0.99 1.07 0.86 1.10 0.99 0.98For area 3, the results are somewhat different, suggesting that in this area the risk of burglarycommunicates over longer distances and endures a little longer.Table 2.19: Weekly Knox analysis for area 3 (values in bold are statistically significantaccording to the residual scores (not shown), N=357)WEEKS1.00 2.00 3.00 4.00 5.00 6.00 7.00 8.00 9.00 10.00 11.00 12.00100 2.20 1.26 1.45 1.14 1.03 1.09 1.03 1.03 1.03 1.06 1.26 0.88200 1.61 1.33 1.39 1.07 1.04 0.99 0.83 0.97 0.75 0.97 0.86 0.72300 1.42 1.51 1.04 1.01 1.17 1.00 0.96 1.03 0.75 0.85 1.02 0.81400 1.33 1.13 1.17 1.03 1.07 1.02 1.07 1.04 1.01 0.86 0.94 0.91500 1.04 1.22 1.02 1.11 0.94 0.89 0.79 0.91 0.92 0.88 0.89 0.93600 1.16 1.17 1.09 1.01 0.88 1.02 0.99 0.94 0.96 1.01 0.82 1.09700 1.13 1.06 0.81 1.05 1.04 0.97 1.00 0.90 1.04 1.01 0.87 0.92800 1.14 1.09 0.97 0.92 0.91 1.09 1.01 0.90 1.01 0.72 0.99 0.89900 1.19 1.06 1.13 1.07 0.95 1.10 0.93 0.98 0.99 1.00 1.00 1.021000 1.17 0.92 1.08 0.91 1.01 0.88 1.03 1.06 1.00 0.93 0.96 0.96For area 4, there appears to be a more classic effect, with the communication of risk being up toaround 400m and for up to three weeks. Again, the risk to those closest to burgled homes is striking.Table 2.20; Weekly Knox analysis for area 4 (values in bold are statistically significantaccording to the residual scores (not shown), N=407)WEEKS1.00 2.00 3.00 4.00 5.00 6.00 7.00 8.00 9.00 10.00 11.00 12.00100 2.12 1.61 1.29 1.00 0.81 1.08 1.38 1.00 1.16 0.84 1.17 1.00200 1.36 1.10 1.36 1.13 1.31 1.08 1.03 0.95 1.04 1.00 0.85 0.87300 1.30 1.18 1.03 1.15 1.14 1.10 1.04 1.13 1.05 1.12 1.05 1.08400 1.22 1.04 1.23 1.13 1.12 1.23 1.15 0.85 0.90 1.05 1.21 1.16500 1.03 1.03 1.03 1.10 0.98 1.12 1.16 1.10 0.98 0.92 0.91 0.87600 1.01 1.01 1.17 1.00 1.05 1.02 1.10 1.14 1.04 0.98 0.84 1.05700 1.07 1.15 1.19 1.08 1.17 1.05 0.98 0.97 1.13 1.04 0.91 0.91800 1.16 1.07 1.06 1.05 0.97 1.07 0.96 1.04 1.06 0.97 0.96 0.98900 1.14 1.21 1.05 1.10 1.28 1.10 1.01 0.86 0.91 0.94 1.12 0.991000 1.11 0.95 0.95 1.01 1.14 1.08 1.03 0.93 1.09 0.97 1.03 0.8616

The results for area 5 again demonstrate that the apparent risk to houses near to the burgledhome is particularly dramatic. However, in common with area 3 the risk of burglary appearsto communicate over a greater distance in this area.Table 2.21: Weekly Knox analysis for area 5 (values in bold are statistically significantaccording to the residual scores (not shown), N=311)WEEKS1.00 2.00 3.00 4.00 5.00 6.00 7.00 8.00 9.00 10.00 11.00 12.00100 5.29 1.14 0.86 0.71 0.71 0.67 0.83 0.29 0.71 1.17 0.50 0.33200 2.31 1.77 1.14 0.77 0.69 1.08 0.92 0.85 0.46 1.38 1.00 0.45300 2.20 1.21 0.80 0.86 1.07 0.62 0.36 1.14 0.64 1.29 1.67 0.92400 2.33 1.07 0.87 0.71 1.21 0.29 0.71 0.71 0.86 1.00 1.15 0.83500 1.64 1.50 1.43 0.69 0.86 0.54 1.38 1.15 0.77 1.08 0.58 1.08600 1.11 1.06 0.72 0.76 0.94 1.00 1.56 0.88 1.29 1.12 0.53 0.73700 1.53 0.72 0.63 0.89 0.72 0.88 1.24 0.72 1.00 1.12 0.38 1.06800 1.47 0.68 0.79 0.83 1.17 0.94 0.82 1.06 1.06 0.56 1.44 0.94900 1.32 1.11 1.32 0.89 1.00 1.06 1.59 1.06 0.78 0.89 0.81 0.631000 0.72 1.06 1.22 1.41 0.89 1.18 1.41 1.47 0.71 1.12 1.40 1.07Thus, although caution is required when interpreting the above results due to the samplesizes involved, the results suggest different profiles across the different areas. That said, forevery area, the largest Knox ratio (and residual) was for the shortest space-time interval,suggesting the ubiquity of this finding at all levels of geographic resolution.Temporal variation?For a number of reasons, it is possible that the profile of the communication of risk for eacharea may change over time. For instance, different offenders may adopt distinct foragingstrategies. Some may prefer to commit crimes very close to previously burgled homesimmediately afterwards, whereas others may prefer to commit them nearby after a shortinterval has elapsed. This may particularly affect the time-space patterning of burglary if newoffenders move into an area, were arrested or move elsewhere. Also, it is possible that thedevelopment or demolition of housing would change the availability of targets and hence thepatterns of crime.For this reason, Knox profiles were generated for Mansfield and ‘A’ Division for sequentialtime periods to determine whether or not the profiles do change, and if so in what way. To dothis, for each area a Knox profile was generated using twelve months worth of data. Next, anew profile was generated using twelve months of data but which, compared to the previousanalysis, included the next two weeks of data, and excluded the oldest two weeks of data.This was repeated 22 times for each area. Thus the analysis was completed using a movingwindow of data, which included new data on each iteration. The results of the analyses werethen animated and compared.The general pattern of results suggested that while the Knox profiles did vary a little from oneprofile to the next they tended to remain stable over time. This may suggest that, in general,offenders adopt similar foraging patterns or that a sufficiently small proportion of activeburglars were ever in custody to modify the overall picture. There is ubiquity of the findingthat burglary clusters in space and time, as evidenced by recent research that shows thatthese patterns are apparent in all other countries for which data have so far been analysedsuch as the UK, US, Netherlands, New Zealand and Australia (Johnson et al., 2006). Oneexplanation for the differences in the profiles observed for each area could relate todifferences in target density between the different areas. Thus, the risk of victimisation maycommunicate over greater distances in areas in which houses are more dispersed, whereasrisks may communicate over shorter distances where houses are spatially concentrated and17

hence available nearby targets plentiful. Considerations of land use provide the mostplausible accounts of such area variation in pattern. Other possibilities exist.Predicting the futureHaving demonstrated that the risk of burglary is indeed communicable in the East Midlands,the next task was to determine whether it is sufficiently predictable to suggest that theproduction of predictive mapping software would aid operational policing. Thus, research wasconducted to compare the effectiveness of:1. Promap;2. what the police are currently doing;3. a simple variant of retrospective hotspotting; and4. what might be expected if areas were prioritised for crime reductive attentionrandomly.At the time of the research, none of the areas currently used retrospective mappingtechniques per se. Instead, they generated pin maps on a fortnightly basis using two weeks’data. This simply involved generating a map which shows the locations of all burglaries thatoccurred within the previous two weeks. On the basis of these maps, operational resourcesmay be deployed. This is similar to retrospective hotspot mapping in that both methodssummarise historic data, and hence in what follows the authors will use retrospectivehotspotting as a proxy for what the analysts currently do. However, it should be noted thatproblems with pin mapping are well documented (e.g. Chainey & Ratcliffe, 2005) and thus theretrospective approach used here as an analogue is significantly superior to what the analystswere doing.Item 2 in the list above has thus been eliminated, being a sub-optimal type of retrospectivehotspotting. As is intuitively clear, random allocation of resources will be less efficient thaneither prospective or retrospective hotspotting, so the choice of method comes down toprospective or retrospective. While necessarily technical, the remainder of this section shouldbe read at least in a cursory way by the interested practitioner, since it identifies thecomputational differences between the two primary contending approaches.The technique most commonly used to identify hotspots involves the generation of a twodimensionallattice to represent the area of interest. As shown in Figure 2.2, a twodimensionallattice (Fig. 2.2(2)) is overlain upon a study area (Fig. 2.2(1)). This comprises aseries of (x*y) cells, each with identical proportions. The challenge of delineating a hotspotlies in the derivation of a set of values, one per cell, that reflects the intensity of crime risk ateach location. Thus, a methodology and mathematical algorithm is required that can generaterisk intensity values for every cell. One technique commonly used to do this is called the‘moving window’. Here, a circle with a predetermined radius (referred to as the bandwidth) isdrawn from the midpoint of each cell (Fig. 2.2(3)), and each of the events that falls within thecircle is used to generate the risk intensity value for that cell. The risk intensity value for eachcell is determined by the number of crime events (in Figure 2.2(1), four for the cellconsidered) that occurred within the circle and how far away they are located from themidpoint of the cell. Those closest to the midpoint are typically assigned a greater weightingthan those further away. To illustrate the method, three of the cells in Figure 2.2(3) have beenshaded to indicate the intensity of risk at those locations. Those shaded darkest exhibit thehighest risks. An example of a hotspot (quartic) function (Bailey & Gatrell, 1995) is describedby equation (1):223 ⎛ d ⎞( ) = 1 ⎟2 ⎜ −iλτs ∑(1)2di≤τπτ ⎝ τ ⎠Where, λτ(s)= risk intensity value for cell s τ = bandwidthd i = distance of each point (i) within the bandwidth from the centroid of the cell18

The bandwidth used to generate the hotspot, and the mathematical equation used togenerate the risk intensity values vary, but the basic rationale is the same. Unfortunately,these are typically not informed by theory or by an in-depth understanding of the crimeproblem, but often because they produce elegant maps. In other cases, the default settingsof the software used are adopted.Figure 2.2: Two-dimensional and three-dimensional hotspot lattices (1. study area, 2.study plus lattice, 3. lattice and retrospective moving window, 4. lattice and prospectivemoving window) (Figure taken from Johnson et al., 2005)(1) (2)(3)(4)d 1Spatial distanceSpatial distanced 2d 1Spatial distanceSpatial distancetimeCrime events Bandwidths Street networks▼ occurred 1 week ago d 1 – Spatial bandwidth● occurred 2-4 weeks ago d 2 – Temporal bandwidthProspective maps are generated using a variant of the moving window technique. The novelfeature is that the amount of time elapsed between events is considered as well as thedistance between the crime events and cells. When defining the model used, considerationof the size of the cells used in the two-dimensional lattice is required.Whilst the optimal size is difficult to determine, it is clear that if a cell is too large much effortwould be wasted in policing low risk locations. Overly large cells would also suffer from theModifiable Areal Unit Problem (Openshaw, 1995) discussed above. For instance, if onecalculates a risk intensity value for one large area which encapsulates two smaller areas withvery different risks, the true risks for neither of the smaller areas will be accurately reflectedby the risk intensity value for the aggregated area. Thus, it is wise to use a cell size thatenables differences in relative risk across (and within) cells within the lattice to be revealedaccurately. However, it is also wise to avoid cell sizes that are simply too small. Forinstance, it is unlikely that a method which used one million cells, each 1m x 1m, would revealmore useful intelligence than a method in which larger cells were used.19

A further point of note concerns the trade-off between having cells of small proportions, whichincreases the precision of the maps generated, and the time taken to complete thecalculations. For instance, the generation of a forecast for a grid which has 100m x 100mcells will take one-quarter of the time that it would take to complete the same task for a gridwhich contains cells that are 50m x 50m. Where analyses are required for an area the size of‘A’ Division, this can be of considerable importance. One issue then concerns the timelyproduction of the forecasts, a consideration whose importance will be amplified if the analysthas to produce forecasts within a limited window of time.The next issue concerns the bandwidth used. As is illustrated in Figure 2.2(4), Promap usestwo bandwidths. The first, the spatial bandwidth, may be calibrated in a variety of ways butshould relate to the distance over which the risk of victimisation is communicable.The second type of bandwidth concerns the time elapsed since a crime occurred and theproduction of the forecast. This bandwidth is conditional upon the first, as a burglary eventshould contribute to the risk intensity value of a cell only if it occurred within a given distanceof that cell. As with the spatial parameter, for the temporal bandwidth a variety of settingscould be used but again this should reflect the period of time over which the communicabilityof risk can reasonably be expected to endure, or slightly longer.A further methodological difference between the authors’ approach and that used inretrospective hotspotting lies in avoiding use of the distance from the midpoint of the cell andthe relevant burglary events in the derivation of the risk intensity values. Consider that forretrospective hotspotting a risk intensity value is calculated for the midpoint of the cell and thisvalue is then allocated to all other points within that cell. If risk intensity values werecomputed for all points within a cell it is unlikely that they would be identical. Thus, using thevalue for the centroid of the cell gives the impression that the risk of victimisation is uniformacross the cell. This is unlikely to be true. This problem will be amplified as the cell sizeincreases. Instead of using the exact Euclidian distance between all events and the cellmidpoint, for Promap the number of cells, the actual unit of analysis considered, between theevent and the cell is instead used. Thus, if a crime occurred within the cell underconsideration, the distance would be zero (actually, for computational reasons 1), if itoccurred within an adjacent cell, two, and so on. By adopting this approach the risk intensityvalues for each cell are likely to reflect more accurately the risks across the entire cell, ratherthan at one single point. Formula (2) was used to derive the risk intensity values for theprospective map, as follows:⎛ 1 ⎞ 1λτ( s)= ∑⎜1+ ∗ci ei c⎟(2)≤τ∩ ≤υ⎝ i ⎠ eiWhere,τ(s)bandwidthc i = number of cells between each point (i) within the bandwidth and the celle i = time elapsed for each point (i) within the temporal bandwidthλ = risk intensity value for cell s τ = spatial bandwidth υ = temporalEvaluating the accuracy of the different methodsAs discussed in earlier work (e.g. Bowers et al., 2004), there has been little attention given tothe measurement of the accuracy of mapping techniques used to inform the deployment ofoperational resources. This is clearly essential. For this reason, in earlier papers the authorsproposed a number of different standard metrics. These include the hit rate, which is simplythe number of future crimes correctly identified. A second important factor is the area of the‘at risk’ locations to be policed. Where operational resources are limited, to compare twotechniques one may need to equate the latter to allow a like-for-like comparison. One way ofcontrolling for this is to select the same geographic area for each method and then comparethe hit rate for each. Where possible this approach has been adopted in what follows. Arelated question concerns the area that should be selected within which crime could occur. In20

a densely populated urban area this can often be done by simply generating a rectangulargrid and selecting a percentage of the cells. However, in a more rural area such asDerbyshire and the other areas considered here, this approach would be less useful. A largeproportion of the grid would not contain any houses, and hence opportunities for burglary.Thus, to make the analyses realistic, for each area an opportunity surface was approximatedusing the data available. Specifically, every burglary was mapped using a GIS and buffer of1km for each point generated. Next, those cells defined by this process were selected andthe rest discarded. Figure 2.3 shows an example for ‘A’ Division. The left panel shows arectangular grid which encapsulates all of the crime events. The panel on the right shows theopportunity surface as defined using the above method. This map (non-white areas) containsapproximately 50 per cent of the rectangle and thus demonstrates the importance of thismethod. For instance, consider that if ten per cent of the surface area of each map wasselected to allow a standard comparison between two mapping techniques, for the map onthe left one would actually select around twice as much area as one would for the map on theright. Whilst this is not a perfect solution it represents a considerable advantage over thealternative by excluding places where burglary cannot happen.Figure 2.3: Opportunity surface for ‘A’ DivisionIn the current research two further standards for comparison were constructed. The firstsimulated what would be expected on the basis of chance. That is to say, how much betterwould it be to use the methods than to simply direct resources to locations on the basis ofchance? To do this, Monte-Carlo simulation was used to generate a chance distribution. Toelaborate, for each area ten per cent of the opportunity surface for that study area wasselected at random using a pseudo-random number generator. The accuracy of theserandom forecasts was then compared to the actual distribution of crimes in the same way asabove. For each area, this was repeated a number of times (in this case, 50) and theaverage of the simulations computed. The size of the cells selected using this approach werevaried (100m, 400m, and 1km) to test the sensitivity of so doing. This is a simpleapproximation to choosing a specific location, a street or a neighbourhood. However, theresults produced using each of these different levels of resolution were remarkably consistentand hence we will report only those generated for 100m cells.A second issue is discussed in some detail below but uses existing statistical methodology toderive a measure of the extent to which hotspots generated by different methods vary interms of coalescence. That is, are there a large number of hot cells that are generally somedistance from each other, or are there a smaller number of coherent hotspots. The latter may21

etrospective method for seven-day forecasts for Mansfield and Corby, although the actualdifferences were (reliable but) small.Table 2.22: Average number of crimes correctly identified per forecast for cumulativemethods (N=22)Retrospective Promap (specific) Promap (classic)2 days 7 days 2 days 7 days 2 days 7 daysWellingborough 3.26 10.96 4.32* 15.68* 4.17* 14.7*Mansfield 3.32 10.18 3.59 11.05* 3.46 10.73*Corby 0.68 3.32 0.90 3.59 0.68 3.55+‘A’ Division 3.32 11.36 4.77* 15.77* 5.55* 19.68** Significantly better than retrospective method (p

Table 2.24: Average number of crimes correctly identified per forecast for single pointmethods (N=22)Retrospective Promap (specific) Promap (classic)2 days 7 days 2 days 7 days 2 days 7 daysWellingborough 3.26 10.96 3.46* 13.00* 3.73* 13.54*Mansfield 3.32 10.18 2.68 8.23 3.32 9.50Corby 0.68 3.27 0.55 2.81 0.91+ 3.46‘A’ Division 3.32 11.36 4.72* 15.68* 5.32* 19.00** Significantly better than retrospective method (p

Figure 2.4: Differences in cells identified as being at the highest future risk byretrospective and prospective methods (cells shaded darkest have the highest riskintensity values)On the one hand this is not a problem as it demonstrates that the prospective method is moreaccurate than the retrospective approach probably because it accurately identifies morelocations that are at risk of burglary, particularly those that are yet to be victimised. Thus, theabove results provide a good test of and support for the theory proposed. On the other hand,in terms of practical policing, as the prospective maps identify more areas this means that theresults are not strictly comparable.One way of facilitating a direct comparison would be to select a smaller percentage of thecells as identified at risk by the prospective method. However, this means that the authorswould unfairly disadvantage the prospective method by constraining how it works in a waywhich is precluded for the retrospective method. An alternative approach is to increase thenumber of cells the retrospective method identifies as having a higher risk intensity value. Todo this, the above analyses were repeated for ‘A’ Division using additional historic data in thegeneration of each forecast. Instead of eight weeks of data, twelve were used. The samevolume of data was also used for the prospective methods to make the test comparable. Dueto the time involved in the analysis, this was not completed for the other areas.Table 2.26: Predictive accuracy for analyses for which the same number of cells wereidentified by each method (N=22)Retrospective Promap specific Promap (classic)2 days 7 days 2 days 7 days 2 days 7 days‘A’ Division 58% 61% 66%+ 66%* 64%+ 64%** Significantly better than retrospective method (p

Patrolling efficiencyAs discussed elsewhere (Bowers et al., 2004), even though a map may predict a largevolume of crime, it may be of little utility in an operational context if it is made of a largenumber of dispersed hotspots. The best map would perhaps be one with a relatively smallnumber of clearly defined hotspots. One way of measuring this is to count the number ofhotspots generated. Another way, which is slightly more sophisticated is to conduct a nearestneighbour analysis.Nearest neighbour analysis is a test of spatial randomness. The nearest neighbour index(nni) in particular measures how clustered points or cells are relative to what would beexpected on the basis of chance. Here, the authors consider the application of this to theanalysis of hot cells - the ten per cent of cells with the highest risk intensity values. How closetogether are they? In this context, a value of one would indicate that the hot cells wererandomly distributed across the area. The lower the value of the nni, the more the hot cellscoalesce to form coherent (and hence policeable) hotspots. The index can be computed forthe nearest high-risk neighbour for each cell, which will often be the adjacent cell. It can alsobe computed for the next nearest neighbour, the next, and so on, up to k-orders. The value ofk is specified by the researcher. Thus, the k-order parameter describes which neighbour isbeing analysed, the nearest (1st order), the next nearest (2nd order) and so on (up to the kthorder). To illustrate, consider Figure 2.5. The nni for the two examples would be the samefor the first order nearest neighbours. However, for the data on the left the second order nniwould be lower than that for the data on the right, thereby indicating that the hot cells in theformer are more spatially clustered than the latter. The most efficient hotspots would perhapsbe those with a low nni for the nearest neighbour, for the next, but particularly for the higherorders. This is because the greater the number of orders for which the nni remains low is anindication of a lower number of hotspots that are more coalescent. A series of dispersedhotspots would have a low distance for the first neighbour, but the distance would increase foreach successive order. A patrolling police officer would then have to spend more time movingthrough low risk areas.Figure 2.5: Illustration of a simple nearest neighbour analysis for two data sets● ● ● ●●●●●Readers may be aware that this type of analysis is traditionally used to examine the degree towhich crime is clustered in space, but as should be evident from the above rationale it is ofclear analytic value here, albeit a novel application of the test. To recapitulate, this type ofanalysis can be used as one index of patrolling efficiency. The lower the nni for higher k-orders, the more efficient the map in this respect. Figure 2.6 shows an example analysis ofthis kind for ‘A’ Division for both a retrospective hotspot and for a prospective map. Theresults are clear. The prospective map has a low nni across all orders, whereas for theretrospective map whilst the nni is initially low, at around order ten it starts to increase.Analyses for other maps generated for different days revealed the same pattern of results. Itis difficult to overstate the importance of this result for the applicability of Promap to patroldeployment.26

Figure 2.6: Nearest neighbour index: retrospective and prospective methods0.60.5nearest neighbour index0.40.30.20.1retrospectiveprospective01 9 17 25 33 41 49 57 65 73 81 89 97k-orderThe results of the nearest neighbour analysis suggest that the prospective method generateshotspots that are perhaps of more practical use than their retrospective equivalents, in thatthey yield more burglary risk per distance moved for patrolling officers.Shift by shift analysisA still further development of the system would be to generate predictions that were specificto particular police shifts. Recorded crime data are imperfect for this purpose, since theyretain a degree of uncertainty about the precise time of an event, but within the limits of thedata to hand, knowing how crime patterns vary by shift offers a clear operational benefit.Otherwise, a forecast for the day would represent the average pattern over the three shifts,but fail to represent the actual geographical patterns for any individual shift. Simpson’sparadox emerges again! To illustrate, consider two areas, A and B. Area A has a high crimerisk during the morning and evening, whereas area B has its highest risk during the afternoon.If one takes the average risks across the two areas, area A would be identified as having thegreater overall risk. However, during the afternoon it would be area B to which operationalresources would most profitably be deployed. The importance of this effect depends upon anumber of factors, one of which is the particular operational tactics to be deployed. Wherehigh visibility policing is used, or temporary measures are implemented with celerity, it iscrucial to know how crime problems move over the course of the day. If longer term crimeprevention interventions are deployed, the highest implementation dosage would be directedto the areas that have the overall highest risks. The point can be made the other way round.Between-shift differences in presenting crime problems should be one factor in determiningthe balance between short-term and long-term interventions.To summarise the story so far:1. the effects identified in Promap are robust, occurring with slight variation across allthe areas studied;2. exploring the relationships closely offers a route forward to refining and optimisingPromap;3. the basic instrument could be developed to meet the tailored needs of a willing policearea.27

3. Tactical options and selecting a pilot siteIn this section there will be a discussion of process by which a pilot area was chosen.Additionally, consideration is given to the types of tactical option that might be usefullyemployed alongside the system developed. The reason for this was that the success of anycrime reduction strategy is contingent upon the identification of the right tactical options thatwill trigger the relevant crime reduction mechanisms (e.g. Tilley, 1993). In the current project,which places its emphasis on the application of a predictive mapping system updatedroutinely, consideration of a number of factors is required. At the time of the research, thefollowing section was produced to provide a number of tactical options that could be usedoperationally to realise the advantages which may accrue from use of the Promap technique.Since the Promap exercise currently concentrates on the crime of domestic burglary(although it has wider potential applicability) so too do the tactical options. It is important tosay that despite this exercise the authors deferred to the policing craft of the operationalofficers who used Promap. The attempt here is simply to make available their current thinkingon tactical implications based on the literature on effective crime reduction.To provide a context for the review that follows, similarities and differences among the threepolice BCUs that were selected as potential pilot sites for the study will be outlined. Followingfrom this, a matrix of a number of tactical options, constructed from evidence of what works,will be discussed. The matrix also provides some guidance on the agencies that are in alllikelihood most suitably placed to implement the different options at the local level. This will befollowed by a discussion of a number of novel possible tactics. Finally, the area andtimescales for the delivery of these tactics will be mentioned emphasising the necessarymechanisms needed to reduce burglary.Potential pilot sitesTo identify a suitable pilot site researchers visited three areas during the first quarter of 2005to gain an in-depth knowledge of how each dealt from day to day with its problem ofresidential burglary from both analytical and operational perspectives. Visits took place overten weeks and involved meetings with key officers as well as attendance at tasking and coordinationmeetings for each of the areas. The three sites visited were Corby andWellingborough in the Northamptonshire force area and, ‘A’ Division in DerbyshireConstabulary. Table 3.1 provides a comparison of the areas and summarises some of themain points for each.All three areas were National Intelligence Model (NIM) compliant and tasking and coordinationis thus organised in a similar way. ‘A’ Division is much larger than either Corby orWellingborough Sectors, comprising an entire BCU, and had a much higher volume ofdomestic burglary. Intelligence is disseminated in a similar fashion across the three areas andall analysts were capable of producing crime point maps using a GIS, and did so fortnightly.None of the analysts in any of the areas routinely (if at all) generated retrospective hotspotmaps of the kind discussed in the previous chapter. All three areas have gazetteers that cangeocode data. This is done only when a map is produced. The geocoding process itselfappeared to be quick in all areas. However if the datum was not assigned a postcode (as isthe case in about 10% of crimes), geocoding was manual and hence time-consuming. Whileeach area had a different crime recording system, ‘A’ Division incorporated the most recentdata as its system was updated every twenty-four hours.28

Table 3.1: Comparison of three potential pilot sitesKey QuestionsHow many burglariesoccur per year onaverage in recent years?What is the annualburglary rate per 1,000householdsWhat is the annualdetection rate?AreaCorby Wellingborough ‘A’ DivisionApprox 900* Approx 650+ Approx 18006.7** 9.9** 7.5***Approx 12%* Approx 10%+ Approx 13%Burglary a priority? Moderate Moderate HighIs there a targetedOperation Bustedburglary initiative in(Jan – Mar 2005)place?How quickly does dataget onto the crimerecording system?Operation Yarn(Feb 2005onwards)24 hrs-2 days 24 hrs-3 days 24 hrsHow many analysts arethere?2 2 2GIS software used? MapInfo MapInfo Blue8What kind of maps doanalysts produce andhow often are theyproduced?- retrospective pointmaps fortnightly- choropleth mapsevery 6 months- retrospectivepoint mapsfortnightly- choroplethmaps every 6months- retrospectivepoint mapsfortnightly- contour hotspotmaps every 6monthsHow often do key officersmeet to discussintelligence and tactics?Are there any waysintelligence isdisseminated to officersoutside of scheduledmeetings?How eager is the area toparticipate?- daily tasking- daily ‘gold’meeting- 2-week tacticaltasking and coordination- 6-monthstrategicassessment‘front page’internal website- daily tasking- 2-week tacticaltasking and coordination- monthly operationalperformance group- 6-month strategicassessmentNo- daily tasking- daily command- 2-week tacticaltasking and coordination- 6-monthstrategicassessmentNoModerately Moderately Extremely* This figure is for the whole of the Northern Area (2 Sectors), thus figures for Corby Sector itself would be lower+ This figure is for the whole of the Eastern Area (2 Sectors), thus figures for Wellingborough Sector itself would belower** Denominator derived from the 2001 Census*** This figure is based on figures from the 2003/2004 business plan29

Selecting a pilot siteThe decision as to which area in which to implement the pilot was taken by a steering groupwhich included members of the Home Office, Government Office for the East Midlands andresearchers from the UCL Jill Dando Institute of Crime Science. The main reasons forselecting the site chosen, ‘A’ Division, were that the Command Team expressed a strongdesire to participate in the pilot, they were especially concerned with the residential burglaryproblem in their Division, and they had been experiencing a stable volume of burglary prior tothe pilot.A tactical options matrix for reducing burglaryInnovations require new ways of thinking. Early software often sought to mimic electronicallywhat had previously been done manually. Only with time was the potential to do things in newways realised. One poignant instance of how innovation requires rethinking comes with theintroduction of the battle tank. Its qualified success upon introduction at the Battle of Cambraiderived from the failure to modify infantry movements to capitalise on the advantages createdby the tank, with long-term unhappy consequences for military tactics (see Dixon 1976). Theworst fate for Promap (short of neglect) would be implementation without rethinking policingtactics.As noted above, prior to the start of the pilot a literature review was undertaken to summarisethe available literature on tactical options used in the reduction of burglary, so that those thatmight be used in response to the maps could be identified. However, by outlining theapparent efficacy of extant burglary reduction tactics, in a sense one falls into the mode ofthought which Promap is intended to make obsolete. Improving one’s capacity to predictthereby changes the terrain. It may mean that hitherto untried or apparently unsuccessfulapproaches become potentially effective. For example, the potential of patrolling as a crimereductive measure has generally been considered low, but when directed by better prediction,this may change. At the risk of appearing to oversell Promap, it may change the ground rulesfor crime reduction. For this reason, it was strongly emphasised that the learning processinvolved in field testing may involve rethinking some of what is set out below.There is no single menu of burglary reductive options that is guaranteed to work in allcircumstances, rather measures need to be tailored according to the specific opportunitiesand situations apparent in an area. One of the main reasons why replications of interventionsoften fail is because the knowledge and understanding of the specific contexts andmechanisms for an intervention are lacking (Tilley, 1993). Sometimes sheer lack of effortmakes for implementation failure. Something that has been successful in one area or situationwill not automatically be successful in others. As such, it must be recognised that whichevertactical options are selected to be used with the predictive mapping system, they may need tobe specifically tailored for the purpose of reducing residential burglary in the chosen pilotarea. Moreover, the most effective burglary prevention strategies involve a combination ofcomplementary responses (Lamm Weisel, 2002), and hence consideration of how a variety ofdifferent interventions might interact, and/or complement the system should be considered.To inform this element of the project, a review of the available literature was conducted and ispresented here as a Tactical Option Matrix in Table 3.2. The matrix can be broken down intofive main component parts: the type of intervention, a summary of the evidence of itseffectiveness, cost (financial and latency of implementation), the geographical coverage of themeasures, and the potential partners who may be involved in the implementation of thatintervention. The following sections elaborate upon and clarify the content of these fiveelements of the matrix.Types of intervention and evidence of effectivenessThe list is not exhaustive by any means, however, the most commonly used crime preventionmeasures for reducing residential burglary are discussed. Over the last two decades therehas been a plethora of research on the effectiveness of various types of crime preventionmeasures and some of the key studies are mentioned here. While the authors are cognisant30

that each study varies in its research design and analytical techniques and thus cannottherefore be directly compared, the Tactical Options Matrix merely provides a summary ofinformation and is therefore not inclusive of every opinion on each measure. In relation to thesymbols used, +, --- and – symbols are used to illustrate whether the intervention has workedin the past, had no impact, or did not work respectively. In some cases, reviews are mixedand both a + and – sign are given to the intervention. As mentioned previously, ultimately thetactical options selected for this project could have included a combination of the interventionspresented here, and would be affected by the partners involved and their ability to implementthe tactics in a timely manner, and the resources available.CostIn the matrix, the cost of implementing successful crime reduction tactics is classified in twoways: firstly in terms of how much the measure will cost to implement in a financial sense;and secondly, how quickly it can be mobilised. Both costs are expressed as either low,medium or high. While a certain measure might be expensive initially, it is essential to look atits cost in terms of swiftness of effect. If an intervention can be implemented quickly and inthose areas identified as being most at future risk, the potential exists not only to reduceburglary but also to prevent low crime areas at a tipping point from evolving into high crimeareas. Thus, it is crucial to compare both types of cost when deciding what will most likelywork best for the selected pilot site and in combination with the proposed approach. Perhapsthe optimal solution is offered by the intervention that has a medium financial cost butswiftness because it can be mobilised quickly.Geographic coverage of the measuresSituational methods of crime prevention work in a variety of ways: increasing effort; increasingrisks; reducing rewards; reducing provocation; and removing excuses. Once the nature of thecrime opportunities is identified, then a selection of measures can be made either to block orreduce these opportunities (Clarke, 1997), and these can be applied on at least twogeographic levels: the specific household or the area. Accordingly, each tactical optionoutlined in Table 3.2 is classified as representing either an area-level or household-specificcrime prevention measure.Measures that prevent crime at an area level will obviously do so by influencing risk at thehousehold level, but they may cost more, take longer to implement and will devote resourcesunnecessarily to many households that are individually at low risk. Conversely, measures thatare extremely effective at the specific household level may have a limited impact across thelarger neighbourhood and thus fail to reduce crime on a larger scale if they are focused ononly a subset of those individually at risk. The ultimate option perhaps is to have acombination of both specific and area-level situational measures that are both successful inreducing crime, providing benefits that address specific problems at particularly vulnerablehouseholds, lower the risk of burglary more generally and produce a realisable deterrenteffect.Partners involved and local responsibilitiesThe involvement of police, local partners and crime reduction agencies are (in currentfashion) important to the success of any crime prevention intervention. Multi-agency crimereduction has been a political priority for some years and, since the Crime and Disorder Act1998 the locally responsible agency is the Crime and Disorder Reduction Partnership (CDRP)rather than the police alone. Table 3.2 shows potential partners such as the police, localauthority and other organisations who could have participated in the implementation of thelisted interventions.Putting together the best tactical options menu can be a difficult process, as the decisiondepends on the available resources as well as the eagerness of police and local partners toget involved. Thus, the aim of the exercise was to inform the discussions with the differentcrime reduction agencies that followed.31

Table 3.2: Tactical options matrixType ofinterventionStudyTargethardening(victim-centred)Tilley & Webb(1994)CCTVGill et al (2005);Welsh &Farringdon(2002); Clarke(1997)RedeployableCCTV (RCCTV)Gill et al. (2005);Gill & Spriggs(2005)High visibilitypoliceBlake & Coupe(2001)Street closuresWagner (1997);White (1990);Beavon,Brantingham &Brantingham(1994)Evidence Cost CostDoes itwork?+/ --- / -++/--++DetailsReduces the risk ofindividual revictimisationbut may not on its ownaffect area rates.Only likely to impact uponcommercial burglariesand those houses withinor very near the areas ofcoverage. Can increasedetectability, act as adeterrent and providereassurance.Limited research suggeststhat while RCCTV allowsflexibility, the system isalso difficult to use andvery sensitive to misuse.No long-term reduction incrime levels.Two-officer patrols havefew advantages oversingle-officer patrols anduse much moreresources.These studies haveshown that there is arelationship betweenstreet access and crimerates.Financiallow/med/highCelerityslow/med/swiftSituationalcrimepreventionArea orspecificlocationPolicePartners InvolvedLocalAuthorityMultiagencygroupLow-high Swift Specific High Med Area High Swift/med Area Med Swift Area Low Slow Area 32

Type ofinterventionSecure byDesign,ProtectingProperty andtargethardeningbyareaAlley-gatingPropertymarkingStudyEkblom (2002);Clarke, (1997);Tilley & Webb(1994)Bowers et al.(2003)Clarke (1997);Laycock (1985)Nelson etal.(2002); Sutton(1998)Evidence Cost CostDoes itwork?+/ --- / -+++/----DetailsDesigning out crime is awell established conceptthat has had profoundeffects on reducing crime.The concept of securelydesigned housing andenvironments is often notdifficult and can beachieved, at the pre-buildstage, for very little extracost.There is good evidenceon the potential benefitsfrom gating rear alleys asa means of reducingburglary.This measure reduces theanticipated rewards ofcrime by making propertyharder to dispose of.However, any impact islikely to be due toassociated publicity notthe intervention itself.These studies emphasisethat in reality, propertymarking does little todeter offenders frombreaking into a house orstealing marked propertyfrom within it.Financiallow/med/highCelerityslow/med/swiftSituationalcrimepreventionArea orspecificlocationLow-med Med Specific &areaPolicePartners involvedLocalAuthorityMultiagencygroup High Med-slow Area Med Swift-Med Specific 33

Type ofinterventionStudyUse ofintelligence &targetingknownoffendersFarrell et al.(1998);Stockdale &Gresham (1995)PublicitycampaignsStockdale &Gresham(1995); Burrows& Heal (1980);Riley (1980);Johnson &Bowers (2003),see also Smithet al. (2002)RepeatVictimisationStrategiesPease (1998);Forrester et al.(1988); Farrell(2005)Ringmaster Lister et al.(2004)Evidence Cost CostDoes itwork?+/ --- / -+DetailsThese studies show thatthe greater use ofintelligence helped focusresources.Financiallow/med/highCelerityslow/med/swiftSituationalcrimepreventionArea orspecificlocationHigh Swift-med Area &specificPolicePartners involvedLocalAuthorityMultiagencygroup +/---Campaigns on their ownrarely change behaviours,although they mayachieve a ‘drip feed’ effectif continued over time.Publicity associated withcrime prevention tacticscan have a crimereductive effect of its own.+ Once a property has been+/-burgled its chances ofsubsequent victimisationincreases; a gradedresponse model (bronze,silver, gold) has provedeffective.A system that alerts localresidents and voluntarygroups of up-to-datecrime information. Whilemany organisationsworried about raising fearof crime among theirclients, they circulated theinfo only to front-line staff.Mixed reviews about thesuccess of thedissemination ofinformation.34Med Swift -med Area Med Med Specific Low Med Area

Type ofInterventionNeighbourhoodWatchSchemes &“CocoonWatches”Forensic Traps(i.e. chemicallytreated mats topick upintruders footprints) & silentalarmsStudyLaycock & Tilley(1995);Forrester et al.(1990)Research todate is limited –Anderson et al.(1995)Does itwork?+/ --- / -Evidence Cost CostDetails+/--- NW has a greater impactwhen residents are homeduring the day. The virtualcocoon that is formed byalert neighbours around theburgled home can increasethe likelihood that anoffender will be caught ifhe/she returns to theproperty. Target cocoonsrather than generalschemes perhaps workbest.--- In relation to burglary ofretail premises both Clarke(2002) and Tilley & Hopkins(1998) emphasise thatthese high-tech devicescan pose numerouspractical problems.Evidence is good on theeffectiveness of silentalarms (Anderson et al.).Financiallow/med/highCelerityslow/med/swiftSituationalCrimePreventionArea orSpecificlocationPolicePartners InvolvedLocalAuthorityMulti-AgencyGroupLow Med Area High Med Specific 35

Other potential tactical optionsAt the time of completing this phase of the research, the published literature on theeffectiveness of various burglary reduction options/tactics was reviewed. To stimulatealternative ways of thinking, also discussed were tentative suggestions for further tacticswhich may complement the ‘near repeat’ burglary victimisation pattern. Due to thecommunicability of risk in time as well as space, meaning that risks surround burgled homesfor a fairly limited time period, many of the suggested measures take existing technologiesand add a time dimension to them. Note that the redeployable CCTV option outlined in thematrix above, an example of this type of approach, is not without its problems. Theseapproaches would therefore only work in a setting where measures could be relocated swiftly.The proposed measures use the following research findings:• that risks are concentrated within 400m and one month;• that some areas have particularly high levels of near repeats; and• that near repeats often share MO characteristics (see Bowers and Johnson,2005b)Use of ANPR/surveillance ‘rings’One possible detection-focused scheme could ‘net’ high risk areas. This could involve amobile ring of covert cameras and/or Automatic Number Plate Recognition (ANPR) devicessurrounding the area. Here a burgled house in the middle of areas where burglary clusters inspace and time would be identified and cameras placed around it at a 400m radius. Thisprovides a way of monitoring all entrances and exits to the area for a limited period of time.This is now set up to act a little like a sting operation as the likelihood is that there will befurther activity in the area. If a further burglary does happen it should be possible to narrowdown when it occurred, to a reasonable time band at least, by asking the victim and looking atthe police record. It would then be possible to look at people and vehicles exiting/entering thearea for a feasible time band during which they could have entered the area and conductedthe burglary. With ANPR it would be possible to work with the Driver & Vehicle LicensingAgency (DVLA) to rule out cars that were registered within the area, which would leave asubset of cars to investigate.Moveable publicityThis would involve the use of offender-orientated publicity. To increase the perceived risks ofcommitting burglary in an area when risk is highest (for perhaps two weeks after an initialburglary), signs could be erected around the boundary of the area with a message indicatingthat people are entering a high-priority burglary clampdown area. The use of surplus policevehicles parked in strategic places (as is sometimes done to deter filling station drive-offs)would be one available form of publicity.Repeat Victimisation neighbours schemeSimilar to a cocoon-watch approach, this would involve always target-hardening houses acertain number of doors away from a repeatedly burgled home, as well as enhancing thesecurity of the property itself. It may be that a general publicity campaign describing the nearrepeatsphenomenon would sensitise otherwise complacent neighbours to take crimereduction measures.Targeted, MO-specific crime prevention adviceThe same vulnerability is often exploited in near-repeat incidents. This means that it ispossible to produce intelligence on the likely MOs of future incidents of burglary. This couldbe used as a basis for prioritising different elements of risk assessments. Crime preventionadvice and assistance would be tailored to MO type and provided to as many nearneighbours of a burgled house as possible. It would (of course) also direct patrolling officersto the part of a home through which entry is most likely to be gained.36

Mobile phone cell broadcastCell broadcast is a very rapid way of transmitting information to all of the mobile phonehandsets within a particular area of radius. Here, a broadcast would go out to everyone in a400m radius of a burgled home as soon as an event occurred. The message would tellpeople to be vigilant as there was a potential risk of burglary. If an offender within the areaowned (or had stolen!) a mobile phone, he/she would also receive the message with apotential deterrent effect.Full forensic examination of transient hotspotsNear-repeats are taken to be more often the work of the same offenders than more isolatedevents, and hence more likely to be prolific and persistent. Thoroughness of forensicexamination should be informed by location within a prospectively identified hotspot.There are strengths, weaknesses and controversies associated with each of the variousoptions suggested above. Furthermore, there is no published evidence on the effectiveness ofthese techniques (although this does not distinguish them from much of what is currentlyattempted). The central point is that Promap requires reconsideration of policing tactics, notsuperimposition on them.Emerging versus enduring risks?Finally, one element of Promap as originally conceived, yet to be discussed, is the facility todistinguish between areas that are currently at a heightened risk of burglary that have beenfor some time (areas with enduring risks), and those that are currently at an elevated risk butthat have not been in the recent past (areas with emerging or transient risk). In those areaswhich represent enduring hotspots, situational crime prevention measures aimed at reducingopportunities for burglary may be the best core option (e.g. alley-gating). On the other hand,in areas that are currently emerging as risky, those interventions that can be implementedmost swiftly would be favoured. This may include targeted police patrols, the use of publicityor deployable CCTV, to name but a few. However, at this stage of the research the CommandTeam and their partners felt that this facility would complicate the pilot and requested that thiselement of the system be developed at a later stage after the pilot, short in duration as it was,had run its course.37

4. System development and evolutionBetween the initial stage of the research and implementation the system was developed in anumber of ways. First, an opportunity surface, a rendering of the spatial distribution andconcentration of houses, was generated and used to weight the predictions generated. AsTable 4.1 reveals, this improved the predictive accuracy of the system considerably. A furtheranalysis showed that for a smaller proportion of the study area (5 per cent) the final versionaccurately predicted the location of around 60 per cent of burglaries, the retrospectiveapproach just over 40 per cent. Expressed another way, the final version was able to identifythe same amount of crime as the retrospective method, but for a patrolling area of half thesize.Table 4.1 Accuracy of the prospective model including the opportunity surface (N=22)Retrospective Promap(specific)Promap (opportunitysurface)2 days 7 days 2 days 7 days 2 days 7 days‘A’ Division 58% 61% 66%+ 66%* 70%* 78%** Significantly better than retrospective method (p

Time of day consistency?As noted above, a further question raised was whether burglaries that occur very close toeach other in space and time (within a few days) are committed at the same time of dayusually within the same police shift as each other? The fact that hotspots change according toshift (e.g. see Ratcliffe 2002) inclines one to that view, but a more direct test was sought. Anaffirmative set of results would suggest that a burglary not only confers risk for a particulargeography and duration, but also for a specific time of day (morning, afternoon or evening).To illustrate, consider that an offender might have a preference for certain activities during theday (see Rengert and Wasilchick, 2000). Consequently, he/she may visit certain locationsduring daytime hours. Whilst there, he/she may commit offences if the opportunity presentsitself. This rhythm of activity, if regular enough, means that an offender will be at certainplaces at certain times. If he/she chooses to commit crimes, and in particular near-repeats atthese locations, then one would expect to see some consistency in the time of day at whichburglaries are committed in those areas. Burglars would appear to work shifts.To examine this hypothesis, one year’s data (January-December 2004) were analysed for thepolice force area (N=8,968). Each crime was compared to every other and the number ofcrimes that occurred at different spatial and temporal intervals identified. In line with theauthors’ earlier work, the spatial intervals here used were multiples of 100m, and the temporalintervals one week periods. Next, all events were compared to see if they occurred during thesame police shift (7am to 3pm, 3pm to 10pm, or 10pm to 7am) and a contingency tablepopulated. Comparisons between events that occurred on the same day as each other wereexcluded from the analyses presented as their inclusion has the potential to inflate theconsistency observed 3 - as a crime series committed during one evening would naturally beconsistent in terms of the police shift during which the events occurred as well as where theyoccurred.To determine during which shift an event occurred, data concerned with the day and time ofeach burglary were analysed. Typically, most burglaries occur when a victim is away from theproperty. Accordingly, rather than recording a single time at which a crime may have takenplace, the police ask for a likely time window, expressed as the earliest time to the latest timethe burglary could have occurred. This was mentioned earlier as a qualification on theanalysis of crimes by shift. In the analysis that follows, the midpoint of these two times wasused as an indicator of the time of the event. Crimes were excluded from the analysis if thetime window (for the earliest and latest day and times) exceeded 15 hours. Follow-upanalyses used shorter intervals of four and eight hours, and revealed the same pattern ofresults.To determine whether the emergent patterns differed from what would be expected on thebasis of chance, if the time of day that near-repeat burglaries were committed were unrelated,Monte-Carlo simulation was used to generate a chance distribution. To do this, using apseudo-random number generator, each crime was randomly assigned the police shift for adifferent burglary (each shift was reassigned only once), and a new contingency tablederived. This was completed 999 times. If the observed results represent a statisticallysignificant pattern, then one would expect that for any space-time combination (e.g. eventsthat occurred within 100m and seven days of each other) the number of burglaries for whichthe shifts are concordant would be greater than the Monte-Carlo results for at least 95 percent of the simulations. This equates to a threshold of statistical significance at the five percent level. In this analysis, as so many comparisons were made, the more conservative oneper cent level of statistical significance was adopted.A subset of the results, shown as Figure 4.1, are presented as the ratio of the observednumber of burglary pairs for which the events occurred during the same time of day (shift),divided by the mean of the Monte-Carlo simulations. Thus, a value of one would indicate thatthe observed value was equal to that expected. Statistically significant results are highlightedwith exaggerated markers on the graph. The dotted line shows that for events which occurredbetween 1,000m of each other, the observed number of burglary pairs that occurred during3 Analyses which included such comparisons produced an identical pattern of results.39

the same time of day as each other was roughly equivalent to what would be expected on thebasis of chance. This likelihood remains stable irrespective of the number of days betweenevents. In contrast, the time of day at which near repeats (those within 100m for illustration)and repeats proper are committed appears to synchronise with antecedent events with anelevated consistency. This pattern was also evident for events that occurred slightly furtheraway (up to around 400m) for one week after an initial event (for clarity of presentation theseresults are not shown). Generally, the distance over which, and the extent to which,consistency was evident for burglary pairs resembled the pattern observed for thecommunication of risk in space and time alone. Thus, there was a pattern of distance decay -events that occurred closest to each other in space and time tended to be committed duringthe same time of day with higher likelihood. This would, therefore, suggest that the time ofday that an initial event is committed offers some additional predictive value beyond thedimensions (days elapsed and distance) from an earlier offence.Figure 4.1: Similarity in time of day for near-repeats and unrelated burglariesObserved/Expected Ratio2.001.901.801.701.601.501.401.301.201.101.000.900.800.700.600.500.400.300.200.100.00RV100m1000m7 14 21 28 35 42 49 56 63 70 77 84 91Days between eventsp

The implication of the results is that the generation of three daily forecasts, one for eachpolice shift, should improve the efficiency of resources deployed. Thus, as a final change tothe system, the predictive model was developed so that it produced different maps for thedifferent shifts, weighting more heavily those events that occurred during the same shift asthat for which the prediction was made.ConclusionThe final Promap system was considerably more accurate than extant methods. The systemwas improved ‘on the hoof’ in a number of ways, partly in response to requests from thepolice Command Team. Risking tedious repetition, it must be stressed that the role ofpredictive mapping in crime reduction is emergent. It provides a tool which may modify policeand Crime and Disorder Reduction Partnerships’ (CDRPs’) practice. There may be falsestarts and initiatives that turn out to be ill-founded. A parallel may be made with cervicalscreening and PSA (Prostate Specific Antigen) testing for cervical and prostate cancerrespectively. Neither of these approaches affords perfect prediction, but both allow areconsideration of risk assessment and treatment prioritisation for the conditions concerned.The argument is that crime reduction is enabled by a more precise specification of risk, butany benefits are contingent upon the shaping of the craft of policing to take maximumadvantage of improved prediction.To give the reader an idea of what the GUI looked like, an example screen image is shown asFigure 4.2. As can be seen, the GUI was quite simple but provided some flexibility so that theanalysts could generate different kinds of maps. Appendix 4 provides further detail of thesoftware interface and illustrates how simple it is to use.Figure 4.2: An example image of the final GUI© Crown Copyright. All rights reserved. Derbyshire Constabulary 100021015.In the Chapters that follow, an account is given of the field trial of the system. Chapter fivediscusses what the police did and problems encountered. To anticipate the findings,41

unfortunately, implementation or use of the system was less than desired. As will becomeapparent, this was not because the utility of the system was perceived to be limited (in factthe reverse was true), problems associated with using the system or even a lack ofenthusiasm, but instead largely because policing priorities changed over the evaluationperiod. This was unavoidable and simply due to the fact that the rate of vehicle crimeincreased whilst burglary declined.42

5. Process evaluationThe aim of this element of the evaluation was to understand and assess exactly what happened andto identify any factors that particularly facilitated or impeded implementation. For instance, whogenerated the maps, how often, how were they interpreted, who looked at them, were tactics shapedby the results, and if so, how?The reasons for this emphasis are (at least) two-fold. First, this type of analysis is essential if themechanisms through which an intervention worked (or failed to work) are to be identified andunderstood. Given that this project is the first in the UK to see if and how police officers can use crimeforecasting methods in an operational context, this was of particular importance. Consequently, thereis much to be learned not only in terms of whether these types of systems could help reduce crime butalso in terms of how they are received by those who would use them, technical issues that may arise,and how the information can be usefully disseminated and interrogated. A particularly poor outcomeof the project would have been to find that the system was not used but with no account of why thatwas the case. Another undesirable outcome would have been to find that the system had not beenused for reasons that could have been corrected during the implementation phase. The work reportedin this section sought to preclude such outcomes.The second reason for conducting the process evaluation was to document how implementationoccurred, who was involved and when, so that intelligent replication elsewhere would be plausible, orso that mistakes made could be avoided in subsequent projects.Central questions addressed whether or not the police and their crime reduction partners used themaps in the deployment of operational resources and, if so, how? Of course, if the answer to the firstof these questions was ‘no’ then it would not be possible to attribute any crime reductive effects to theapproach. Thus, the process evaluation was integral to, and informed the evaluation of change inlevels of crime that followed. In subsequent sections the approach taken for this element of theevaluation is considered and results discussed.Process evaluation methodologyA variety of measures were used to identify and document the processes through which the pilot wasimplemented. These included: semi-structured interviews, a survey of front-line police officers, thecompletion of a tactical options log by the Command Team, and direct observation by the researchteam. In this section, to provide an overview of the approach taken, each of the different methodsused will first be discussed before discussing the collective results.The maps were routinely generated by the two crime Intelligence Analysts who worked on theDivision. To gain an understanding of how they perceived the usefulness of the system, both in termsof user friendliness of the software and the maps generated, three semi-structured interviews wereconducted with both analysts at the beginning, middle and end of the seven-month implementationperiod. The questions asked were divided into two general categories: 1) how useful did they feel themaps were; and, 2) how much did the production of the maps impact on their daily work.Semi-structured interviews were also carried out with 12 Section Sergeants in Alfreton, Belper, LongEaton and Ripley towards the end of the pilot period. These were intended to provide an in-depthnarrative of how useful, if at all, the maps were to those who used them, and to seek understanding ofhow they were used on a day-to-day basis.At the end of the implementation period, a survey was conducted with 57 front-line police officers in ‘A’Division to help gain an understanding of how useful the officers felt the maps were as well as whether43

any tactical options were employed as a result of the maps. The questionnaires consisted mainly ofclosed questions, however, there were three open-ended questions which were intended to helpprovide a more complete picture of officers’ feelings towards the maps. Officers were provided with apaper copy of the questionnaire and reassured that their responses would remain anonymous and notbe provided to the Command Team.To ensure an objective account of how the maps were used on a day-to-day basis, the CommandTeam were asked to fill out a log sheet each time they used the maps. This meant that the tacticaloptions selected by the Command Team as a result of the maps were logged and evidenced on aregular basis.In addition to asking crime analysts and police officers about the project, researchers observed dailybriefing meetings and tactical assessment and co-ordination meetings on three occasions at thebeginning, middle and end of the implementation period. It was originally hoped that these visits couldbe done on a surprise basis to minimise the extent to which demand characteristics came into play.This proved impossible due to the co-ordination and compliance of key members of the CommandTeam and Intelligence Analysts who required warning of each visit due to their busy schedules.Rather than present the results collected using each methodology separately, to avoid repetition, asynthetic approach is taken to help provide a complete picture of how the pilot was implemented, usedand received in ‘A’ Division.‘A’ Division, management and day-to-day runningDerbyshire Constabulary is divided into four Divisions: Alfreton ‘A’, Buxton ‘B’, Chesterfield ‘C’ andDerby ‘D’. With a population of 240,000, ‘A’ Division covers an area of approximately 150 square milesand comprises a number of large towns as well as more remote rural communities. The CommandTeam comprises a small number of key officers including the Chief Superintendent andSuperintendent for the Division, who are in charge of all Division-level decisions. In addition, the FieldIntelligence Unit and Volume Crime Team also help develop intelligence and carry out essential policeoperations and make tactical decisions at a Division level.‘A’ Division is divided into five sections, which comprise a total of 32 beats (with between five andseven in each section). Each section has its own Criminal Investigation Department (CID), LocalIntelligence Officer (LIO), Uniformed Officers, and Community Support Officers (CSOs). Pivotal tothese smaller areas are the Section Inspectors and Sergeants who have complete ownership overtheir section in terms of tasking and co-ordinating their section’s beat team. Each beat has a numberof Beat Officers who are responsible for front-line police activity and providing reassurance tomembers of the community.As already discussed, ‘A’ Division is National Intelligence Model compliant.By following standards set out by NIM, ‘A’ Division manages intelligence using a structured system ofmeetings and briefing documents to ensure the dissemination of knowledge reaches all of the keypeople and is done in an efficient way:• Daily Briefings (daily): short meetings are held at ‘A’ Division Headquarters, daily at 9:15a.m. andare chaired by a member of the Command Team and held via videoconferencing across all fivesections. The meetings are used to discuss current problems and target resources for the dayacross the whole Division.• Section Tasking Meetings (daily, shift by shift): occur on a shift by shift basis (7am, 3pm and11pm) and are chaired by the Section Sergeant in charge of that shift. These meetings areintended to be a quick and efficient way to inform and task police officers using an intelligencedocument prepared by the section’s Local Intelligence Officer.44

• Tactical Assessment and Co-ordination (fortnightly): tasking and co-ordination is an importanttime when the intelligence produced by analysts is disseminated among key officers. Thesemeetings occur on a fortnightly basis and are attended by the Tasking and Co-ordination Group(including key officers such as the Chief Inspector, Section Inspectors, Community Safety Teammembers, and Intelligence Analysts). Chaired by a member of the Command Team, alldepartments and partnership agencies are invited. These meetings are primarily informed by theTactical Assessment document prepared by analysts. This document contains details ofsignificant crime trends, observations and current issues (intelligence and prevention) on anumber of crimes as set out in the Control Strategy (e.g. violent crime, dwelling burglary, andvehicle crime). The main purpose of these meetings is for key members of the Division to cometogether to discuss pertinent issues that have occurred since the previous meeting and targetavailable resources accordingly.• Strategic Assessment (every 6 months): A comprehensive Strategic Assessment document isproduced by analysts every six months. This includes analyses of any crime series, trends andhotspots that have been identified on ‘A’ Division and is discussed among senior rank officers.IT and disseminationTo minimise disruption to what the police have to do on a day-to-day basis it was important to ensurethat the maps could be generated for, and discussed during, the existing structure of briefingmeetings. During the first months of implementation, as with any pilot, a number of problems arose.These along with the solutions to them are discussed below.How long did it take the analysts to produce the maps each day?It was decided by the Command Team at an early stage that since the two Intelligence Analysts wereusually responsible for the production of Divisional intelligence products (including maps), they werebest placed to produce prospective maps. Due to data protection issues and other security concerns,the force IT department felt that the Promap software could not be installed directly onto thenetworked force computers. Instead the software was installed onto a stand-alone laptop which wouldbe used by the analysts to produce the maps. This was extremely unfortunate as it increased the workinvolved in generating the maps. It meant that the data required to produce the maps would not bereadily available on the computers used. Thus, each time a new map was produced, the analysts hadto do the following:1. extract the data required using a networked machine;2. write this to a CD 4 ;3. copy the data onto the laptop;4. produce the maps;5. copy the maps to a CD;6. open the files on the networked computers, and store them in the folders allocated to eachsection.This, of course, increased the amount of time required to produce the maps. Had the software beeninstalled on the networked machines, steps 2, 3 and 5 would have been redundant. This method ofoperating should be avoided at all costs in any replication as being immensely wasteful of a skilledhuman resource (for a more detailed discussion of these issues, see Appendix 1). At the start of thepilot, the maps took the analysts up to an hour each to produce. As they became more familiar withthe process involved this decreased significantly, ranging from between 15 and 40 minutes towards4 USB drives were disabled on all force machines for security reasons.45

the end of the pilot. In addition to producing the maps, an extra 30 minutes of the analysts’ time wasrequired to enable them to attend the daily meetings held at 9:15a.m. Their attendance at thesemeetings was required to ensure that the maps were understood when being discussed by the seniorofficers, and so that maps could be interrogated in some detail if required.Dissemination of the mapsSome initial concerns arose pertaining to the practicalities of producing and disseminating the maps.These had to be overcome quickly. The main concern was about how to provide the information in apractical format to those who would use it, as the Section Sergeants, who would ultimately have toinspect the maps, were based at different locations.An initial thought was for the Intelligence Analysts to email copies of the maps to the relevant seniorofficer in charge of tactical delivery. This proved problematic as not all officers had routine access toemail. Furthermore, the maps produced required that sufficient memory was available for theirstorage, making it difficult to send them as attachments. Other ideas included the production of hardcopies of the maps which could be attached to regular briefing documents. This would have beencostly and would have limited the number of different maps that could be generated and the timeinvolved in their production.Ultimately, the compromise adopted was for the Intelligence Analysts to produce the maps and thencopy them as jpeg files into each section’s briefing folder, located on the Divisional IT system. Thefolders also contained daily briefing documents prepared by the LIOs that all Section Sergeants wereresponsible for, so this ensured that the maps would not be missed. Maps were transferred to thebriefing folders in this way from the insemination of the pilot, in August 2005.Once the maps were available, they were accessed in two main ways (see Figure 5.1):(1) If burglary was a particular concern for the Division then a member of the Command Team wouldrequest that the maps be shown during the daily briefing meeting at 9:15a.m. where they could directpolice action and tactics to Section Inspectors and/or Sergeants.(2) Section Sergeants who wanted to look at the maps specific to their section could also access themdirectly. During the first few weeks of the pilot, ‘A’ Division invested in a videoconferencing system aswell as plasma screens for each of the five sections. The videoconferencing system was to be usedduring the daily briefing meetings to connect all sections, thus eliminating the need for officers to travelfrom their section to Divisional Headquarters. The plasma screens were installed in each section andthis had the advantage that Section Sergeants could also show the maps in their shift-specific briefingmeetings. This enabled all front-line officers to see the maps and the areas they should be targeting.During the pilot period, the Intelligence Analysts and Command Team discovered a problem with themaps produced for the Section Sergeants. The issue was that when producing the maps, only alimited number could easily be generated for each section due to the time involved, and thus theanalysts had to make a selection based upon their own judgement about what was interesting andwhat was not. Unfortunately, as the software was not installed on the force IT system, the SectionSergeants could not themselves interrogate the maps if they wanted to take a closer look at thepattern of risk within a particular area, or if they wanted to ‘zoom’ into a particular location. Nor couldtime be devoted to taking a closer look at each map for every section during the daily meetings. Theresulting difficulty was that when a particular Section Sergeant felt that burglary was a problem andwanted to look at the maps in some detail, only a partial picture of the problem might be available. Thesolution to this difficulty was to provide a version of the Promap software to each section. Thus, a newversion of the software was developed that could be used by the LIOs in each section. Rather thanhave each section produce their own maps, which would have required them to download new dataeach day and then generate the maps, the analysts continued to process the data with their software.However, instead of just providing the sections with jpeg images of the maps, they were now able to46

export the maps themselves, which could then be interrogated in detail by the Section Sergeantsusing the software provided. Consequently, the Intelligence Analysts trained each LIO to use thesoftware once they had acquired the relevant equipment, in this case laptops. Unfortunately, due tologistical issues involved in acquiring the laptops, this did not take place until the end of December intwo sections and the end of January in the other three. Thus, this solution was not fully realised untilvery late in the pilot period. Figure 5.1 illustrates this method of dissemination.A fourth method of dissemination was through the fortnightly tactical assessment and co-ordinationmeetings. The maps were used during these sessions if, and only if, the Command Team felt thatburglary had been a significant concern over the last two weeks. These meetings allowed for afocused discussion of the maps and potential tactical options that could be used in the identifiedareas. One such response that evolved in January 2006 was that of a collaboration between theCommunity Safety Team who worked with the Division to organise and implement a promap-specifictargeted operation. The operation was intended to provide high visibility policing and crime reductionmeasures to the areas identified through Promap, with the intention of:• increasing public awareness around home security and property marking; and• increasing neighbourhood watch schemes in priority areas.This involved the co-ordination of the Divisional Community Safety Unit and all sections. To elaborate,on days that the maps were produced (i.e. Mondays and Thursdays) one Section Beat Officer and oneCommunity Safety Constable would use the Mobile Police Station (i.e. a marked police van) with theintention of targeting identified hotspots in two ways: firstly by providing a high visibility presence todeter potential offenders; and secondly to increase public awareness visiting residential premises inthe identified areas giving out crime reduction advice, making home security assessments whereneeded, referrals to other agencies and considering whether a neighbourhood watch scheme wouldbe appropriate for the identified area(s). It was felt by ‘A’ Division as well as the research team thatthis collaboration was a success in its own right as it helped improve the already good relationshipbetween the Community Safety Unit, community partners and front-line Beat Officers.47

Figure 5.1: Promap dissemination process across ‘A’ Division (squares shaded grey were present throughout the pilot while squares in white were onlyadded to the dissemination process during the final two months of the pilot)Intelligence AnalystsProduce maps and put them innetwork Divisional Section folders41 2 3Tasking and Co-ordinationIf burglary is of considerableconcern, maps are discussedwith key officers and partnersand operational tactics aredecided uponDaily Meetings (‘daily prayers’)The Command Team decidewhether or not the maps should beshown during the meeting. If theyare, any tactics that evolve from themaps are discussed and plannedSection Sergeants (daily shiftmeetings)Section Sergeants responsiblefor accessing the maps anddeciding whether or not to usethem for targeted police activitySection LIOsResponsible for navigating themaps with the Section laptops.This is dependent on Sectionspecific problem areasCommunity Safety TeamHave the potential to use theinformation discussed around themaps to deliver tactics, e.g. givingcrime prevention advice toresidents in Promap identifiedhotspotsSection Inspectors andSergeantsResponsible for implementingthe tactical options with frontline officersBeat OfficersResponsible for carrying outany tactical options in responseto the maps48

How often were the maps produced?From the beginning of the pilot (15 August 2005) and up until 4 September 2005, the maps wereproduced five times a week (Monday to Friday). However, producing the maps five days a week had anumber of disadvantages. Most important, the areas identified as being most at risk did not alwayschange significantly from day to day. In reality, the pattern from one day to the next would not betotally stable, but would be serially correlated. Thus, some areas identified as being at a high risk onone day would most likely be at high risk the next day, and possibly the day after that. The reason forthis is that the data used to generate the predictions would be similar from one day to the next,reflecting the activity of offenders. The exception would be during periods of time when a large volumeof offences takes place each day. In this case, the predictions would vary considerably from one dayto the next. However, even when the daily volume of crime was low, as a few days pass thepredictions would change, keeping pace with the flux of crime. Nevertheless, it is possible thatviewing the maps one at a time, on sequential days, may have created an illusion of stationarity. Afterall, to detect differences in the maps each day officers would have to remember the exact locationsidentified from one day to the next, and human memory and perceptual systems are known to besusceptible to distortion. To illustrate, a series of predictions are shown in Figure 5.2. Thesepredictions were generated for one area every Monday for a period of four sequential weeks. As canbe seen, in some areas the risks are relatively stable, but elsewhere more fluid. Detecting thechanges, instead of being deceived by an illusion of stationarity, requires one to look quite carefully.For example, the reader is invited to look at the pattern from week to week in the centre of the map.This area always has some degree of risk associated with it, but the areas shaded in the darkestshade (blue for those with intact colour perception and a colour version of the report) clearly move. Infact the blue areas always move at the level of resolution at which policing tactics would be deployed,but a glance at the map may suggest stability. Thought should be given to modes of depiction of mapswhich highlight change.Figure 5.2: A series of predictions for one area (blue areas are those most at risk)Further concerns about generating the maps each day arose because it was felt that they would notbe taken seriously if officers were forced to look at them five days a week, on top of all the otherintelligence that they had to assimilate. Over exposure to the maps might mean that officers wouldlose interest. For these reasons, it was decided that from 5 September 2005 the maps would beproduced three times a week (Monday, Wednesday and Friday). During the next six weeks the projectreceived a lot of positive reception from the officers and seemed to be used regularly. However, fromNovember onwards the residential burglary numbers dropped further and, consequently, officersperceived that the three maps produced each week were beginning to look similar to each other.Consequently, the Command Team decided that at this point the maps would be produced twice aweek (Mondays and Thursdays). If, however, the analysts felt that there was a significant concernregarding residential burglary and that producing the maps more than twice a week would be helpful,they were free to do so, and did on a number of occasions. Maps were produced twice a week from14 November 2005Timing issuesUnfortunately, the pilot got off to a false start because of a number of fundamental timing issues. First,the Divisional Commander, who had been extremely supportive and helpful during the planning stagesof the pilot and who was instrumental in securing ‘A’ Division as the pilot site, received a promotionand left the Division in August. As is the way, there was no advance warning that this would happenand thus appropriate measures could not be taken to minimise the impact this had on the pilot. Due to49

the busy schedule of the new Divisional Commander, the research team was unable to meet with andbrief him on the project until the end of October, nearly two and a half months into the pilot.Furthermore, key members of the Command Team had arranged annual leave and were unavailableduring the early days of the project period. During the first meeting that could be arranged post-August, it was revealed that instead of using the predictive capability of the maps, when the mapswere used, attention was instead focused on the locations of burglaries that had occurred within thelast two weeks, the approach that had previously been used on the Division (and the approach thatPromap outperformed and was designed to replace). The reason for this was that it was felt that theareas identified as being at risk were too large. Consequently, the software was further refined togenerate predictions for smaller areas.As a result of these factors, summarised in Figure 5.3, it took a few months for the system to be finetunedand for the Command Team members to unite to give Promap a high profile across the Division.Because the pilot was only seven months in duration and almost three months had elapsed before themaps were being produced in the way intended, it was not implemented over a sufficient period to fulfilits potential. As will become apparent, the pilot expired just when it was becoming accepted andunderstood across the Division.User-friendliness and impact on workloadGenerally, the Intelligence Analysts found the maps easy to produce, however, they voiced the opinionseveral times throughout the pilot that having the system on the Divisional network could havesignificantly improved the ease of producing the maps. More automation of the process would havealso reduced the time and effort involved. For example, the analysts regularly spent time manuallyremoving distraction burglaries from the dataset (as they felt that these burglaries would conform todifferent patterns and trends than other residential burglaries, although this is an empirical questionworth addressing). Only approximately 80 per cent of the data were automatically geo-coded by theforce IT system, and hence the analysts had to spend time manually geo-coding the remainder. Ofcourse, these issues are germane to the force IT system rather than the Promap software. However,they are worth noting as it is likely that similar issues may arise in other police force areas.Despite these issues, the analysts did not feel that the maps impacted greatly on their day-to-daywork. In some cases, where they were specifically responsible for summarising a burglary problem(e.g. producing a problem profile or a specific tasking document), they reported that the system helpedto illustrate patterns and trends they felt would not otherwise have been considered. The onlysignificant effect that the production of the maps had on the Intelligence Analysts was their startingtime at work. Because the maps had to be available to discuss in the daily briefing meeting at9:15a.m., the analysts had to ensure they were at work by 8a.m. By the end of the pilot, each analystwas only producing the maps once a week, and so the time involved was not a considerableinconvenience.50

Figure 5.3: Timeline for Prospective Mapping Pilot in ‘A’ Division (August 2005 – March 2006)Mid-End Aug• Several of the Command Teammembers on leave in August15 Aug• Official start date5 Oct• Maps stopped being produced 5 times/wk (M-F) and wereproduced 3 times/wk (M, W & F)End Aug• DivisionalCommander JohnWright left ‘A’Division26 Oct• Evaluation visit #1• Meeting to inform new Divisional Commanderof project1 Nov• Analyst interviews #123 Dec• Laptops collectedby Alfreton andBelper (LIOsshown how to usePromap)20 Feb – 3 Mar• A Researcher from the Home Officeinterviewed four Section Sergeants (all butIlkeston)8, 9 Mar• Analyst interview #3 (Deborah Rimell)• Community safety team (90 packs)18 Jan• Evaluation visit #2• Analyst interviews #220 Mar• Community Safety Team (45 packs)28 Feb• Official end date of pilotAug-05 Sep-05 Oct-05 Nov-05 Dec-05 Jan-06 Feb-06 Mar-0617 Aug• Retrospectivepoints shown onmaps28 Oct• Retrospectivepoints nolonger shownon maps14 Nov• Maps produced 2times/wk Mon & Thurs4 Nov Consensus by the steering group thatmaps will now be produced 2 times/wk(Mon and Thurs)51End Jan• Laptopscollected byLong Eaton,Ripley &Ilkeston19 and 23 Jan• Community Safety Team proactive patrol,handing out door-to-door info packs (70and 35 packs handed out)1, 2 Mar• Evaluation visit #3• Analyst interview #3 (Bill Wallage)• Promap survey (57 officers)• Community Safety Team (100 packs)

Tactical deliveryCommand Team daily briefing (9:15am)As described above, prospective maps were shown and discussed at the daily briefingmeetings if the Command Team felt they would be a valuable addition to the meeting. Ifburglary numbers had been sufficiently low for the past few days, then the maps were usuallynot shown. Table 5.1 shows the number of times that prospective maps were used in the dailybriefing meetings from initial implementation to the end of the evaluation period. Numbersdecreased dramatically towards the latter months because burglary numbers were falling soconsiderably, and Divisional priority had been shifted to dealing with auto-crime which roseover the same period.Table 5.1: Number of times prospective maps were used in ‘A’ Division’s daily briefingNumber of times prospectivemaps were used in dailybriefingsAUG 05 6SEP 05 8OCT 05 8NOV 05 4DEC 05 1JAN 06 2FEB 06 1Total 30According to the tactical logs completed by the Command Team, of the 30 times prospectivemaps were used in daily briefings, tactics were employed as a result of the maps 27 timesacross the Division. The most common tactics employed in response to the maps were footpatrols, drive-through patrols in hotspot areas and the targeting of known offenders believedto operate in or nearby the areas identified.Section front line responseOn 2 March 2006 a survey was conducted with 57 officers across all sections to assess theirlevel of knowledge and understanding of Promap as well as to identify any factors that mayhave either facilitated or impeded implementation. A copy of the questionnaire used isprovided as Appendix 2. Table 5.2 provides a breakdown of the sample of officers surveyedacross the Division.52

Table 5.2: Sample characteristicsNumber %SexRankTime in RankSectionMale 42 73.7Female 15 26.3PoliceConstable49 86.0Sergeant 6 10.5Inspector 1 1.8SpecialConstable1 1.8Less than 1year6 10.51-5 years 30 52.6More than 5years20 35.1No answer 1 1.8Alfreton 9 15.8Belper 13 22.8Ilkeston 15 26.3Long Eaton 12 21.1Ripley 8 14.0Table 5.3 shows the number of respondents who had heard of prospective mapping, bysection. In all sections, aside from Ilkeston, the majority of respondents had heard of Promap.In Ilkeston, only four of the fifteen respondents in that section, had heard of the pilot. 5 Thosewho had heard of Promap were asked how they would define Promap in their own words.Two researchers, one who did not work on the project and one who did, examined theresponses to ensure that there was no bias of interpretation. The inter-rater reliability for thetwo researchers was high (Cronbach’s Alpha=0.99) and thus the coding appropriate. 52 percent (21) correctly identified the definition of Promap, i.e. the maps ‘predict where burglariesare likely to occur’. Those that did not provide an accurate definition showed a basicunderstanding of the system (e.g. ‘shows burglary hotspots’ and ‘shows crime hotspots’). Afew respondents, 21 per cent (8) did not fully understand the maps as they thought that theyillustrated both burglary and auto crime hotspots.5 It was suggested by a member of the Command Team that one possibility why the number of respondents is so lowis because of the terminology used in the questionnaires. It seems that the Division was using the term ‘JDI maps’when referring to the pilot, whereas the questionnaire specifically asked if respondents had heard of ‘ProspectiveMapping’. One would assume that if this were the case, however, that the same trends would have been evident inthe other four Sections.53

Table 5.3: Number of respondents who had heard of prospective mapping, by sectionSection Yes No No, b/c newTotalstarterAlfreton 88% (8) 12% (1) 0% (0) 16% (9)Belper 92% (12) 8% (1) 0% (0) 23% (13)Ilkeston 27% (4) 73% (11) 0% (0) 26% (15)LongEaton83% (10) 17% (2) 0% (0) 21% (12)Ripley 75% (6) 12% (1) 12% (1) 14% (8)Total 70.2% (40) 28.1% (16) 1.8% (1) 100.0% (57)When asked how frequently the maps were routinely used for targeted police activity, officersreported using them with different frequencies. Table 5.4 shows that around 30 per centreported using them more than twice a week, and a similar proportion reported that theirsupervisors had used them at this rate. Follow-up interviews with the Section Sergeants, theLIOs and the analysts suggested that this had been the case when burglary was a priority, butthat, understandably, the maps were used significantly less when it was not. The generalperception, which is corroborated by the records from the tactical options log, was that themaps were used more during the first few months of the pilot when burglary was more of apriority 6 , although the maps were still produced and disseminated throughout the pilot periodtwice a week thereafter.Table 5.4: Number of times maps were used for targeted police activityNumber of timesofficers usedprospectivemaps(s) fortargeted policeactivityNumber of times officers’supervisors usedprospective map(s) fortargeted police activityMore than twice a21% 21%weekTwice a week 9% 11%Once a week 11% 11%Once every two4% 4%weeksOnce every three0% 2%weeksLess than once a5% 2%monthDon't know 0% 2%No answer 27% 47%6 Unfortunately, this was when the maps produced reflected retrospective point maps rather than predictive surfaces.54

Table 5.5 shows the number of tactics reported to have been used in response to theprospective maps when they were used. Those most commonly used were drive-throughpatrols in identified areas, foot patrols and the targeting of offenders known to operate in oraround the areas highlighted in the maps.Table 5.5: Number of respondents who were either involved in or responsible foremploying operational tactics, by sectionTactic either employed orinvolved in Alfreton BelperLongEaton Ripley Ilkeston TotalDrive-through patrols in hotspots 6 6 6 4 2 24Foot patrols around hotspots 5 5 1 3 1 15Targeting known offenders 1 3 4 3 0 11Repeat victimisation strategies 0 1 3 1 0 5Target-hardening 1 1 1 1 0 4Publicity campaigns 2 0 0 0 1 3Redeployable CCTV 0 0 1 1 0 2Crime prevention advice given 0 0 1 0 0 1As noted, officers were sometimes tasked to target known offenders, and although this issomething they have always done, it was suggested by some that the prospective maps hadhelped them combine the intelligence gathered on offenders by Intelligence Analysts andLIOs with the identified areas, providing an overall enhanced tactical option for catchingoffenders. In one case, one officer mentioned that “I can’t give names or give numbers but it’s[offenders who have been caught have been linked back to certain areas that were identifiedby the maps] certainly happened a few times”. Repeat victimisation strategies are apparentlycommonly used on ‘A’ Division, regardless of Promap; however, one Section Sergeant inLong Eaton mentioned that he tasked officers to revisit burgled properties to give out crimeprevention advice and this would often be done in areas identified by the maps. Targethardening also is something that is normally done on the ‘A’ Divisions; it is part of theDerbyshire Constabulary business plan. This involves a Crime Prevention Officer visiting localresidents and offering crime prevention advice and advice on having security measures fitted.If the householder wishes to have security measures installed, the crime prevention officercan arrange for this. Although respondents from each of the sections mentioned that theyused target-hardening in response to the maps, when speaking to the Sergeants in each ofthe sections, they mentioned that target-hardening was a tactic that has always been used onthe Division, regardless of Promap and thought that perhaps the respondents had recalledbeing involved in target-hardening prior to filling out the survey and thus had misconstruedusing the tactic as a result of the pilot. This, of course, serves to illustrate the importance oftriangulating evidence from different sources when completing a process evaluation.Although it was decided by the Command Team in October 2006 that it did not wish to launchany publicity campaigns in response to the pilot for fear of frightening local residents, somesections chose to put short press releases in their local paper(s) alerting residents to ‘takecare’ and make sure their property is safe and offering crime prevention advice. However,this form of publicity was not specific to the project, even if it was evoked by it.A final aspect of police working that was not highlighted in the survey, but came up a fewtimes when interviewing Section Sergeants and LIOs, concerned the planning of ‘night-timepatrols’. According to Sergeants in Alfreton, Long Eaton and Ripley, when resources55

permitted, up to four plain clothed officers were deployed at night and patrolled those areasidentified by the prospective maps. This tactic was usually done at night because the nightshift is often slow moving. Consequently, there are fewer calls for Beat Officers to respond tocompared to shifts during other times of the day, which frees up available resources forproactive patrols. Although it was not possible to enumerate the frequency with which nighttimepatrols were informed by the predictions generated, this does illustrate that whenresources were available and burglary was a priority, police officers used Promap for crimeprevention and detection purposes.As previously mentioned, during the final few months of the pilot, and during March (onemonth after the end of the pilot), the Community Safety Unit deployed a mobile police vehiclein the hotspot areas designated by the maps and distributed residential burglary crimeprevention packs door-to-door in the areas identified as being at the most risk. The dates onwhich, and the number of packs distributed were as follows:• 19 January: 70 packs;• 23 January: 35 packs;• 2 March: 100 packs;• 9 March: 90 packs;• 20 March: 45 packs.Usefulness of mapsAs can be seen from Table 5.6, 87.5 per cent of the sample who had heard of the pilot foundthat prospective maps were either easy or fairly easy to interpret, whereas only 12.5 per centfound the maps to be either fairly difficult or difficult to interpret. In terms of how useful officersfound the maps to be, although 67.5 per cent of the sample found the maps to be either veryor at least somewhat useful, 32.5 per cent (10) found them not to be very useful. Similarly,when asked whether or not the maps identified risky areas that respondents would have nototherwise have considered risky, 47.5 per cent (19) of the sample felt that the maps ‘never’identified unknown risky areas, whereas 42.5 per cent (17) felt that the maps ‘sometimes’identified risky areas that they would not have otherwise considered as such. In relation tothis point, research conducted by the authors of this report (McLaughlin et al., 2006)demonstrates that whilst police officers often have a good impression of where burglarygenerally occurs, they are less accurate at identifying where it recently took place. Theimplication is that they are unlikely to be able to anticipate where it will next occur. Moreover,interviews with LIOs and the analysts, suggested that although the maps often identifiedpriority areas in known risky neighbourhoods, the precise location or timing of elevations inrisk were not always expected.Table 5.6: The interpretation and usefulness of prospective mapsNumber %Interpretation of mapsEasy 10 25.0Fairly easy 25 62.5Fairly difficult 4 10.0Difficult 1 2.5Usefulness of mapsVery useful 4 10.0Somewhat useful 23 57.5Not very useful 10 25.0Not useful 3 7.556

In addition to the questions discussed above, survey respondents were asked two openendedquestions to canvass any thoughts not captured by the closed questions asked. Firstly,they were asked to outline any extra comments that they might have about the usefulness ofthe maps and how they might be improved. Secondly, respondents were asked to outlineextra comments about their knowledge and understanding of prospective maps and anytactical options that they had employed. To be frank, on the basis of experience the authorshad expected negative feedback in response to these questions. Instead, few chose torespond. Of those that did, 27.5 per cent (11) commented on how the maps could beimproved, for example three suggested that:“the colour of the hotspots should be red not blue”;one suggested that:“the maps are too simplistic and should show MO, point of entry, class etc.”Perhaps surprisingly only two opined that:“the maps don’t tell me anything that I did not already know”.In terms of extra comments about police officers’ knowledge and understanding of Promap,only four made comments, with one asking“how accurate are the maps?” and another suggesting that“the maps are good for supporting information for patrol strategies”.Although the official end-date of the pilot was the end of February 2006, the maps are stillbeing produced on a biweekly basis to inform tactical delivery. Moreover, the CommandTeam have decided to use the system more frequently if the rate of burglary increases andthe priority returns to this type of crime.SummaryAs already discussed, during the fist two to three months of the pilot, rather than using thepredictive capability of the system developed, pin maps showing the locations of recent crimepatterns were used. This was unfortunate as it was in conflict with what Promap wasdesigned to achieve - accurate predictions of the future locations of burglaries for those as yetunvictimised (as well as repeat victims).Despite this and some initial negativity towards the approach, there now seems to be amoderate to high level of acceptance of the maps and the Division has embraced the systemas a useful tool. Unfortunately, during the pilot period there was a substantial increase in theftfrom vehicles across the Division (and elsewhere in Derbyshire) in October which wascoincident with a reduction in burglary. Consequently, there was a shift in policing priority atthis time towards vehicle crime across the Division. The problem with this in terms of theevaluation was that it meant that less time was focused on burglary, and the application of thePromap system. In short, this precludes a fair evaluation of the impact on crime of Promap.Nevertheless, given that this is the first attempt to implement such a system in the UK, it is anon-trivial finding to be able to say that feedback from those using the system has been onthe whole very positive. In fact, most of those interviewed, formally or otherwise, havesuggested that should burglary increase in the future they would use the system to tackle theproblem, which in itself illustrates the utility of the system as perceived by those who wereexposed to it. In line with this observation, despite the fact that the pilot has come to an endand that burglary continues to remain low within the Division, at present the maps are stillgenerated every two weeks to inform the tasking and co-ordination meetings. A further57

observation made by those who used the system was that when used it helped to focus theirattention on the problem and how they might reduce it.As a further testament to police officer acceptance of the usefulness of the system, a numberasked if it could be used to predict incidents of vehicle crime. Since completing the pilot, it hasbeen shown (Johnson et al., 2006) that patterns of theft from motor vehicle (TFMV) conformto the same spatial and temporal patterns as burglary and hence the answer to this questionis likely to be yes. Thus, an anticipated future development of the system would be to facilitatepredictions of TFMV also.Thus, from the perspective of the evaluation, the pilot served to demonstrate police officeracceptance of, and confidence in, the system, and to show that it could be used in anoperational setting without disrupting other activity. Important lessons were learned in relationto the dissemination of the maps. Initially, it was thought that the analysts could generate anddistribute the maps, with Section Sergeants having only to look at the output. However, it wassoon felt that the sections would benefit from being able to interrogate and navigate the mapsthemselves, thereby allowing them to gain a more detailed picture of the problem. Thesolution to this problem was straightforward, but for a variety of logistical reasonsimplementation of it occurred too late in the evaluation period to have any potential impact oncrime.58

6. Changes in patterns of burglaryKey to robust evaluation is the notion of effect signatures. This concerns the pattern of resultswhich reflect mechanism – just as signatures bespeak identity – and equally important, whichfail to reflect cherished but erroneous ideas about mechanism. With respect to the currentproject, a number of signatures would be anticipated if reductions in crime were achieved as aconsequence of Promap- influenced intervention. For example, reductions in crime would beexpected to coincide with the timing and intensity of implementation. However, changes inimplementation intensity can be measured in a number of ways. In a project such as this,where the deployment of police resources varies by time of day as well as day of the year,analyses should consider variation in crime for different intervals of the day as well as byweek or month of year.With respect to changes in patterns of crime in space, particular a-priori expectations mayalso be expressed. For example, one would expect distinct changes in the spatialconcentration of burglary following intervention. In the extreme, if an intervention had theeffect of preventing all burglaries subject to prediction (those that conform to an identifiedregularity), the spatial distribution of crime would appear random following intervention. Ofcourse, this scenario is unlikely but distinct changes in the spatial concentration of crimeshould be observed where an intervention has an effect.Finally, in addition to expecting changes in the temporal and spatial distribution of crime,distortions in the space-time clustering of crime would be expected. To illustrate, considerthat spatial hotspots of crime are defined by a series of crimes that occur during someinterval. The precise timing of the events is unimportant, with a spatial hotspot being definedonly by virtue of a clustering of events in the spatial dimension. An alternative signature is aseries of crimes that cluster in both space and time; a localised spate. Where policeintervention is designed to anticipate such activity, as was the case here, one indication ofsuccess would be a truncation in the duration of patterns that might suggest such activity.Consequently, a series of novel analytic techniques were developed to detect signatures ofthe type discussed. However, as is illustrated in earlier sections, implementation of the pilotwas insufficient to facilitate an appropriate test of the potential crime reductive impact of thepredictive approach. As such, it is suggested that presentation of a sophisticated statisticalanalysis here would be unwise. Instead, in the sections that follow, simple analyses arepresented to examine the changes in the patterns of burglary observed before and during thepilot project. The reader interested in the more detailed analytic approaches used is referredto Appendix 3 for an illustration of the analyses and a more detailed discussion of theirrationale.The basic approach adopted in most evaluations is to compare the change in the volume ofcrime before and after intervention in both an action and comparison area. If a reduction isobserved in the action but not comparator, or the reduction in the former exceeds that in thelatter then a positive inference may be drawn. Figure 6.1 shows the change in the volume ofburglary in the action area, and a comparison area, Derbyshire ‘C’ Division. The latter wasselected partly because the trends in the two areas followed similar patterns prior tointervention but also because discussions with the Command Team suggested that bothDivisions employed similar approaches to operational policing, not least because they werelocated within the same police force, Derbyshire. It is evident from the time-series graph,which shows the patterns for two years prior to the pilot and thereafter, that there was areduction in burglary in both areas over time. Prior to intervention, which of the two areas hadthe higher volume of burglary each month varied. For this period, the mean monthly count ofcrime was 112 (SD=31.6, N=24) in the pilot area, 116 (SD=38.2, N=24) in the comparator.59

A simple time-series analysis (see Appendix 3) confirmed that the two areas followed asimilar trend and experienced a similar volume of burglary for the two years before the pilot.At the start of the pilot (August-September) the volume of burglary rose in both areas, afterwhich it remained somewhat stable in the comparison area but fell in the pilot area. For themonths of January and February 2006 the volume of burglary in the pilot area was the lowestit had been for at least the last five years, being less than half the volume for the equivalentperiod of time in the previous year. 7 For reference, also shown in Figure 6.1 are the times atwhich the major implementation outputs of the pilot began.At this point, it is perhaps useful to provide the reader with a little more contextual informationregarding other policing initiatives implemented in ‘A’ Division before or around the time of thepilot. Interviews with the Command Team, LIOs and the analysts for the Division, suggestedthat the only intervention implemented across the Division was the prioritisation of prolific andpriority offenders (PPOs). This began around February 2005 and is ongoing. The aim of theintervention is to target prolific offenders, those who commit the bulk of offences, across theentire BCU with the aim of detecting and consequently reducing crime.Given the focus of this intervention, it is plausible that this could have impacted upon theincidence of burglary before and during the pilot period. To see if any changes in burglaryoffences observed in ‘A’ Division were likely to be attributed to this strategy, the number ofdetections recorded for the seven-month periods before and during the pilot were consideredand compared to those in ‘C’ Division (which also focused on PPOs). This analysis revealedthat in ‘A’ Division the number of detections per 1,000 burglaries increased slightly over time,but less so than it did in ‘C’ Division for the same period of time. Moreover, in the pilot areathe rate of detections was a little lower for both periods than it was for the same period of timeone year earlier, whereas for the comparison area the reverse was true. This pattern ofresults would suggest that changes in the pilot area over time are unlikely to be attributable tothe targeting of prolific offenders.Complicated analyses could be conducted to attempt to determine whether the reduction inthe incidence of burglary observed was statistically significant. Readers interested in whatsuch analyses might show are directed to Appendix 3 of this report. However, as alreadydiscussed it is proposed that the interpretation of such analyses would be unclear asimplementation of the pilot on the ground was so limited. Thus, in this section a simplemeasure is presented as a guide to the changes observed. The metric computed, an oddsratio, merely contrasts the change in the intervention and comparison areas before and afterintervention. An odds ratio of one indicates that the changes in the two areas werecommensurate, suggesting no change in the pilot area. An odds ratio of greater (less) thanone suggests a reduction (increase) in the intervention area relative to the change observedin the comparison area. The statistical significance of the odds ratio can also be computed(see Lipsey and Wilson, 2001) by estimating the standard error of the value derived.This technique, which is readily interpretable, has been frequently used in researchconcerned with what works in reducing crime (for examples, see Welsh and Farrington, 2006;Gill and Spriggs, 2005), but is not without it critics, particularly for analyses conducted at thesmall area level (for which fluctuations over time may occur even in the absence ofintervention: Marchant, 2005). However, the problems articulated about this approach arelikely to be less problematic for analyses conducted at the BCU level, for which the variationover time is less of an issue than for smaller areas (see Farrington and Welsh, 2006). Thus,the approach is used here because it provides a simple assessment of how things changed inthe pilot area relative to the comparator.7 Perhaps ironically, it was at this point in time, during which the burglary rate had remained stable for the last year,that ‘A’ Division was selected as the pilot location.60

61MonthAug-03Sep-03Oct-03Nov-03Dec-03Jan-04Feb-04Mar-04Apr-04May-04Jun-04Jul-04Aug-04Sep-04Oct-04Nov-04Dec-04Jan-05050Burglary count100150200250Figure 6.1: Time-series graph of the count of burglary before and during pilotFeb-05Mar-05Apr-05May-05Jun-05Jul-05Aug-05Sep-05Oct-05Nov-05Dec-05Burglary preventionpack distributionTargeted patrolsControlPilotJan-06Feb-06

Two approaches were used to compute the standard errors (since these are critical in determining thesignificance of the effect-size derived), one used by Farrington and colleagues (see Welsh andFarrington, 2006), the other by Gill and Spriggs (2005), 8 Both approaches converged on similarestimates and hence only the former are reported here.To calculate the odds ratios (OR), the count of crime for the same period of time before and during thepilot were contrasted. 9 The standard errors were computed in the usual way (see Lipsey and Wilson,2001) as well as using monthly variation as suggested by Gill and Spriggs (2005). Table 6.1 showsthe count of burglary for the periods before and during the pilot phase along with the OR, andassociated confidence intervals and z-score. The confidence intervals shown, calculated using thetraditional approach indicates the upper and lower estimates of the odds ratios. These suggest thatthe true odds ratio lies somewhere between 0.99 and 1.34. The z-score provides an indication of thelikely statistical significance of the OR. For a two-tailed test, the z-score is required to exceed 1.96.On the basis of these results, the analysis would suggest that the reduction in burglary observed in ‘A’Division exceeded that in the comparison area, but that the trend was marginally non-significant.Table 6.1: Change in the volume of burglary and odds-ratio statisticsBefore After GrosschangeOddsratioConfidenceintervalz-scorePilot 663 554 109 1.15 0.99-1.34 1.80Comparisonarea684 659 25Given the duration of the pilot and the implementation issues experienced the only surprising featureof this result is that it came so close to statistical reliability. The above analysis considers only generalchanges in burglary over time and thus in the next section changes in the time of day when burglariesoccurred will be examined.Change in the time of day burglaries were committedAs noted in the implementation section of the main report, there was a consensus of opinion that thepredictive maps were more frequently used during the evenings, when reactive policing demands onpatrolling officer time were typically less acute. Thus, one expectation would be that if the predictivemaps were used more frequently for resource allocation during the evenings, there should beobserved a greater reduction in the number of burglaries that occurred during this time of day followingthe inception of the pilot.To explore this, rather than analysing changes in the rate of burglary for each hour of the day, theapproach adopted in an earlier section of the report was used. That is, the shift during which everyburglary occurred was identified (by computing the mid-point of the earliest and latest times the eventcould have occurred) and the monthly patterns summarised. Only burglaries for which the window ofreporting was less than eight hours were included in the analysis. Consequently, there was someattrition in the volume of data analysed (50% of events occurred within an interval of eight hours). Afurther analysis (not shown), conducted using a window of fifteen hours to increase the sample size(64% of events occurred within a fifteen-hour reporting window), revealed the same pattern of results.Figure 6.2 shows the change over time in the proportion of all burglaries committed within ‘A’ Divisionthat occurred during the evening shift, before and after the start of the pilot. A trend line is included toillustrate the pattern prior to the start of the scheme. The figure illustrates that there was some8 Gill and Spriggs (2005) use a slightly different approach to calculate the standard by considering the monthly fluctuation in thevolume of crime to reduce a problem known as over-dispersion.9 The pilot started in the middle of August but in the analyses that follow, for simplicity 1 August is taken as the start date. Theeffect of so doing is to make the analyses more conservative.62

variation in the proportion of burglaries that were committed during the evening (mean = 0.20,SD=0.07, N=52) before the start of the pilot, and that over time the overall trend was ever so slightlyupwards. Following the start of the pilot, and particularly from September onwards, the proportion ofburglaries committed during the evening dropped (mean = .11, SD=0.04, N=7); the monthly variationobserved also decreased. Thus, it would appear that following the start of the pilot the proportion ofburglaries committed during the evening decreased.Figure 6.2: Changes in the proportion of burglaries committed during the evening over time (forevents for which the interval between the earliest and latest reporting times was less than 8 hours)0.400.35BeforeDuringLinear (Before)Proportion of all burglaries0.300.250.200.150.100.050.00MonthApr-03Dec-02Aug-02Apr-02Dec-01Aug-01Apr-01Dec-05Aug-05Apr-05Dec-04Aug-04Apr-04Dec-03Aug-03Further analyses explored the change in the patterns for the other two shifts. Naturally, changeswould be expected in at least one of these as the unit of analysis was a proportional measure.Figures 6.3 and 6.4 show the changes observed during the morning and daytime, respectively. Therewas little change in the mean proportion of events committed during the morning before (Mean=0.39,SD=0.09, N=52) and during the pilot phase (Mean=0.36, SD=0.09, N=7).63

Figure 6.3: Changes in the proportion of burglaries committed during the morning over time(for events for which the interval between the earliest and latest reporting times was less than 8 hours)0.700.60Proportion of all burglaries0.500.400.300.200.10BeforeDuringLinear (Before)0.00MonthDec-05Aug-05Apr-05Apr-03Dec-02Aug-02Apr-02Dec-01Aug-01Apr-01Dec-04Aug-04Apr-04Dec-03Aug-03During the daytime, more events were committed during the pilot phase (Mean=0.53, SD=0.09, N=52)than before (Mean=0.40, SD=0.10, N=7), although (and unlike the trend observed for eventscommitted during the evening) the pattern observed during the pilot phase was not completely unlikethat observed in the recent past.64

Figure 6.4: Changes in the proportion of burglaries committed during the daytime over time (forevents for which the interval between the earliest and latest reporting times was less than 8 hours)0.700.60Proportion of all burglaries0.500.400.300.200.10BeforeDuringLinear (Before)0.00MonthDec-05Aug-05Apr-05Apr-03Dec-02Aug-02Apr-02Dec-01Aug-01Apr-01Dec-04Aug-04Apr-04Dec-03Aug-03Considering the changes in the comparison area, as shown in Figure 6.5 there was also observed areduction in the proportion of burglaries committed during the evening, although the change over timewas not as distinct (means = 0.24 and 0.17, SDs = 0.11 and 0.05, N=52 and 7, respectively) as for thepilot area and in the comparison area the trend was more similar to the historic trend, particularly ifone takes account of the outlying observation in July.As a further analysis, and to take account of seasonality, the proportion of burglaries committed duringthe evening for the pilot interval (August 2004 to February 2005) and for the same period for the yearbefore (August 2003 to February 2004) were compared for both pilot and comparison areas. For thepilot area the proportion was lower during implementation (mean=0.11, SD=0.04, N=7) than for thesame period the year before (mean=0.17, SD=0.07, N=7), a difference which achieved statisticalsignificance (z=2.03, p

Figure 6.5: Changes in the proportion of burglaries committed during the evening over time (forevents for which the interval between the earliest and latest reporting times was less than 8 hours)0.600.50BeforeDuringLinear (Before)Proportion of all burglaries0.400.300.200.100.00MonthApr-03Dec-02Aug-02Apr-02Dec-01Aug-01Apr-01Dec-05Aug-05Apr-05Dec-04Aug-04Apr-04Dec-03Aug-03The above results, which show a selective pattern, are certainly in line with what would be expected ifthe system had been used most frequently during the evening and was helpful in reducing burglary.However, to provide a more conclusive result, data for a longer period of time during whichimplementation occurred would be required to demonstrate continuation of the observed trend. Inparticular, this would further help to rule out any seasonal effect. Thus, it is suggested that the resultshown is enticing but inconclusive.As discussed at the start of this section, detailed analyses were conducted to explore changes overtime in the spatial and spatio-temporal distribution of burglary. For reasons already discussed, theseanalyses are presented in Appendix 3.66

7. ConclusionsIn this section, the main findings of the research will be discussed and recommendations inspired bythem presented. The central aims of the current project were as follows:• to determine whether patterns of burglary are communicable across a range of areas;• to test the accuracy of a predictive mapping system across the same areas and compare itwith contending alternatives;• to tailor the system for use in an operational context;• to see if the system could be used operationally and how it was received by those who mightuse it; and• to test the efficacy of the system during a field trial in one police BCU.The first four aims were achieved. The results demonstrated that across all areas the risk of burglarywas communicable. Following a burglary at one home the risk to those nearby was elevated for aperiod of time afterwards. This pattern conformed to a pattern of spatial and temporal decay. Thosenearest were at the greatest risk, and the change in risk decreased as time elapsed. Further researchdemonstrated that when events occurred close in space and time they tended to do so at a similartime of day. For repeat victimisation proper, this consistency also emerged around five to six weekslater, a finding compatible with explanations of the timing in the elevation in risk often observed for thetime course of repeat victimisation more generally (i.e. offenders revisiting homes to steal replacedgoods). Exploratory analyses described later in the report considered the length of space-timeclusters of crime for one area. This approach will be developed in ongoing research underway by theauthors.Recommendation 1 Given the evident ubiquity of the space-time clustering of burglary across theareas studied, it would be wise for some of the analyses discussed in the report to be conducted bypolice analysts to provide a better understanding of crime in their area. Whilst simple analyticsoftware is currently unavailable, an application that performs the same kinds of analyses describedhere is currently being developed as part of a National Institute for Justice-funded project (Ratcliffe,2006). This will be released as freeware in 2007 and will be compatible with a variety of off-the-shelfsoftware tools.Recommendation 2 Where it is found that crime (burglary and other types) clusters in space and timeacutely, strategies aimed at the prevention of further crimes in a local spate could be developed toprevent or detect crimes. As a short-term strategy, this could be partially achieved (for example) usinga GIS and the prioritisation of police resources by police analysts to homes nearby, and similar tothose recently burgled. A number of police forces, including Cleveland, Dorset and the Police Servicefor Northern Ireland have developed such strategies. Relative to Promap, this will be a sub-optimalapproach but may be a useful first step.As for the accuracy of Promap, this method was shown to be superior to those extant, even when thelatter were optimised. In addition to producing maps that more accurately predicted where futureburglaries occurred, the maps identified more coalescent areas for targeted patrolling. Furtherdevelopment of the system involved the inclusion of an opportunity surface that was used to weightthe predictions made. This enhanced the accuracy of the system still further. Prior to implementation,a final feature of the mapping system developed was the facility to produce predictions on a shift-byshiftbasis, which understandably appealed to those who used the system.Despite initial scepticism by some officers, by the end of the pilot the system was well received andgenerally perceived as a useful tool for targeted crime reduction. As testament to this, a number ofofficers asked if the technique could be applied to other types of crime such as theft from a motorvehicle. Recent work by the authors suggests this to be the case (Johnson et al., 2006), and furtherprojected work will seek to develop this use. An additional feature of the future system, welcomed bythose interviewed as part of the project, would be the facility to anticipate changes in which crime typeshould be prioritised for reduction over the next few days or weeks.67

With respect to implementation realised during the pilot, despite widespread belief in the usefulness ofthe system, a number of factors limited the extent to which operational tactics were deployed inresponse to the predictions generated. These have been described in the body of the report. Theanalyses of the potential impact on burglary, many of which were novel, were carried out and arereported in fulfilment of contract rather than in expectation of success given tardy implementation. Itwas not realistic to anticipate crime reductions during the currency of the project as delayed, and theencouraging trends at its end must be considered as an unexpected bonus. Had they not emerged,the writers would have been no less excited and energised by the potential of prospective mapping forcrime reduction. The reasons for that excitement will be elaborated below.Recommendation 3 Analytic methods used to identify mechanisms of change in the evaluation ofcrime reduction interventions are often limited. The techniques described in Appendix 3 of this reportillustrate a number of ways in which particular ‘signatures’ that might bespeak mechanism forinterventions aimed at dispersing hotspots may be sought. It is recommended that such approachesare used more widely to evaluate intervention.The Promap system trialled in Derby allows the prediction of burglary events far better than previousmapping systems. Predictability of crime location is a major aid to its prevention, by disruption anddetection. Sting operations are uniquely effective because the time and location of crime is known.Prospective mapping in the shape of Promap takes us much closer to achieving predictability. Thequantum leap in performance it achieves over previous systems comes by the incorporation of thetime dimension. The soccer cliché is that a striker has to be in the right place at the right time to scorea goal. The former striker, Gary Lineker, makes the point that this is meaningless in that if a player isin the right place at the wrong time, that makes it the wrong place. If the player is in the wrong place atthe right time, that makes it the wrong time! Only by thinking about time and place together astime=place does the cliché make sense (however banal). Similarly, a police officer must be in the righttime-place to disrupt or detect crime. Historically, and with a few recent exceptions, crime mapping forthe police service has neglected the time dimension. The Derby trial of Promap highlighted how‘slippery’ and shift-specific hotspots are, overlain on a degree of location stability. Perhaps it is themismatch between a conventional map showing a nightclub to be a hotspot with its quietness everyMonday morning as experienced by a patrolling officer which leads to a schizoid view of the relevanceof mapping for operational policing. Promap, by the centrality of time in its construction, negates thatproblem.Some technologies are recognisable as having massive potential future applicability while early indevelopment. (What use is a new born baby)? Nanotechnology is one obvious current example. Stemcell use for organ repair is another. Whilst on a different scale, the writers believe Promap is anotherexample. However, there must be a development process which does not seek dramatic early crimereductions (although some should occur) and be addressed to resolving two issues, set out in thefollowing paragraphs. Should effort be given to resolving them? The writers’ emphatic view is that theyshould. The game is very much worth the candle.The first problem standing between Promap and routine use is that the police have to attend incidentsof crime and disorder generally, while Promap as yet covers a limited number of crime types.Development must extend to all categories of crime and disorder. The relevant science completed, therelative importance of different events in driving patrolling patterns must be incorporated in thePromap algorithm. This is a matter of policing policy. Unpublished work from the Department ofOperational Science at Lancaster University in the late 1970s demonstrated that such policy choicesmust be tested and that police preferences changed according to the mix of offences detected. Apartfrom this immediate complication, police preferences may change from time to time with changingpriorities. There must, in short, be some weighting to direct a patrol to a location which will host threeassaults and two thefts rather than a location which will host three thefts and two assaults.The second problem to be resolved before Promap can realise its potential would remain even afterthe first is resolved. Promap output must be delivered in real time to police officers in the form ofpresumptive patrolling patterns. This is not difficult even with current technology but comes at a cost.The most prominent obstacle is bedding Promap into policing craft. Officers must always be able tooverride a presumptive patrolling pattern on the basis of personal knowledge, but must come to trustthat the presumptive pattern of patrol is soundly based.68

Recommendation 4 Further development of Promap or a variant should, in the authors view, be apriority. The research presented here demonstrates the superiority of the approach over existingcontenders and shows that it is welcomed by the police. To achieve what is clearly possible willrequire further development of the system and a series of field trials across a range of differentcontexts.Recommendation 5 The utility of the approach should be explored for a range of crime types.Recommendation 6 It would be useful to distinguish between areas for which risks are increasingand those for which they are stable or declining. Operational tactics would vary for these two types ofarea.If the writers might be allowed to end on a flight of fancy, they envisage a situation in perhaps fifteenyears when predictive mapping is available for all crime types, real time information on risk is availableto police patrols, where the seriousness of different crime types is weighted automatically so that anoptimal patrolling pattern is provided to each police vehicle to maximise the total seriousness of crimesto be preventively patrolled. Used in concert with Lab-on-a-chip forensic testing, where DNA and othertests would be possible in police vehicles, would facilitate swift forensic identification of perpetrators ofcrimes not prevented, and patrolling informed by Promap would mean faster response times to arrivebefore crime scenes are compromised for forensic purposes. In parallel with optimised patrolling,Promap would deliver information about longer-term patterns and stabilities in crime and disorder toCrime and Disorder Reduction Partnerships, enabling them to put in place design and maintenancechanges. Nothing in such a future is unfeasible even with today’s technology. It does, however requirean effort of imagination to discern the centrality of prospective mapping to such a future.The authors’ nightmare scenario is that Promap suffers death by a thousand trials. When assaultivecrime is shown to be Promap predictable, minor changes in such crime will likely be achieved.However, since the police have to put Promap for assault alongside non-Promap-based decisionmaking for other crime types, they will have to decide at any moment whether they are in Promap ornon-Promap mode. Only when Promap is the default basis for routine policing will its benefits becomevisible. This is a brazen plea that any further funding of Promap focuses on the eventual realisation ofan integrated system, rather than short-term and crime-specific operational trials.69

ReferencesAggresti, A. (1996) An Introduction to Categorical Data Analysis. New York: Wiley.Allison, D.B. and Gorman, B.S. (1993) ‘Calculating Effect Sizes For Meta-Analysis: The Case of TheSingle Case’. Behaviour Research and Therapy, 31(6), 621-631.Anderson, D., Chenery, S. and Pease, K. (1995) ‘Biting Back: Tackling Repeat Burglary and CarCrime’. Crime Detection and Prevention Series, Paper No. 58. London: Home Office PoliceDepartment.Ashton, J., Senioe, B., Broen, I. and Pease, K. (1998) ‘Repeat Victimisation: Offender Accounts’International Journal of Risk, Security and Crime Prevention, (3), 269-280Bailey, T. C. and Gatrell, A. C. (1995) Interactive Spatial Data Analysis, Harlow: Longman.Beavon, D. and Brantingham, P. (1994) ’The Influence of Street Networks on the Patterning ofProperty Offences’. In Crime Prevention Studies, Vol. 2, R. Clarke (ed.), Monsey, N.Y.: CriminalJustice Press.Besag, J. and Diggle, P.J. (1977) ‘Simple Monte Carlo Tests for Spatial Pattern’. Applied Statistics,(26), 327-333.Blake, L. and Coupe, R. (2001) ‘The impact of single and two-officer patrols on catching burglars inthe act’. British Journal of Criminology 41(2), 381-396.Bowers, K.J., Johnson, S.D. and Pease, K. (2004) ‘Prospective Hot-Spotting: The Future of CrimeMapping?’ British Journal of Criminology, 44, 641-658.Bowers, K., Johnson, S. and Hirshfield, A. (2004) ‘Closing off Opportunities for Crime: AnEvaluation of Alley-Gating’. European Journal on Criminal Policy and Research, 10,(4),283-308.Bowers, K.J., and Johnson, S.D. (2005a) ‘A Test of the Boost explanation of Near Repeats’.Western Criminology Review, 5(3),12-24.Bowers, K.J., and Johnson, S.D. (2005b) ‘Domestic burglary repeats and space-time clusters: thedimensions of risk’. European Journal of Criminology, 2(1), 67-92.Burrows, J. and Heal, K. (1980) ‘Police Car Security Campaign’ in Designing Out Crime, Clarke,R.V.G. and Mayhew, P. (eds.). London: HMSO.Chainey, S. and Ratcliffe, J.H. (2005) GIS and Crime Mapping, Federation Press: Sydney.Clarke, R.V. (2002) ‘Burglary of Retail Establishments’. Problem-Oriented Guides for Police SeriesNo. 15.Washington: U.S. Department of Justice.Clarke, R.V. (1997). Situational Crime Prevention: Successful Case Studies, 2nd edition. Guiderland,NY: Harrow and Heston.Clarke, R.V.G. and Eck, J. (2003) Become a Problem-Solving Crime Analyst. Jill Dando Institute ofCrime Science: London.Dixon, N.R.F. (1976) On the Psychology of Military Incompetence. London: Jonathan Cape.Eck, J. and Weisburd, D. (1995) Crime and Place. Monsey NY: Criminal Justice Press.Ekblom, P. (2002). ‘Future Imperfect: Preparing for the Crimes to Come’. Criminal Justice Matters, 46,38-40.70

Ericsson, U. (1995) Straight from the Horse’s Mouth. Forensic Update, 43, 23-25.Everson, S. and Pease, K. (2001) ‘Crime Against the Same Person and Place: Detection,Opportunity and Offender Targeting’. In Crime Prevention Studies Volume 12: Repeat Victimisation,G. Farrell and K. Pease (Eds.) New York: Criminal Justice Press.Everson, S. (2003) ‘Repeat Victimisation and Prolific Offending: Chance or Choice?’ InternationalJournal of Police Science & Management, 5(3),180-194.Farrell, G. (2005) ‘Progress and Prospects in the Prevention of Repeat Victimization’. In theHandbook of Crime Prevention and Community Safety, N. Tilley (Ed.). Cullompton: Willan.Farrell, G. and Pease, K. (1994) ‘Crime Seasonality: Domestic Disputes and Residential Burglary inMerseyside 1988-90’. British Journal of Criminology 34, 487-497.Farrell, G., Phillips, C and Pease. K. (1995) ’Like Taking Candy: Why does repeat victimizationoccur?’ British Journal of Criminology, 35, 3, 384-399.Farrell, G., Chenery, S. and Pease, K. (1998) Consolidating Police Crackdowns: Findings from anAnti-Burglary Project. Police Research Series Paper 113. London: Home Office.Farrington, D.F, and Welsh, B.C. (2006) ‘How Important is “Regression To The Mean” in Area-BasedCrime Prevention Research’. Crime Prevention and Community Safety, 8, 50-60.Forrester, D., Frenz, S, O’Connell, M. and Pease, K. (1990) The Kirkhold Burglary PreventionProject: Phase II. Crime Prevention Unit, Paper 23. London: Home Office.Forrester, D., Chatterton, M., and Pease, K. (1988) The Kirkholt Burglary Prevention Project,Rochdale. Crime Prevention Unit Paper 13. London: Home Office.Gill, M. and Spriggs, A. (2005) Assessing the impact of CCTV. Home Office Research Study No.292. London: Home Office.Gill, P. and Mathews, R. (1994) ’Robbers on robbery: offenders perspectives’. In Crime at Work, M.Gill (Ed.). Leicester: Perpetuity Press.Gill, M, Rose, A. and Collins, K. (2005) A good practice guide for the implementation of redeployableCCTV. Home Office Online Report No.16. London: Home Office.Johnson, S.D., and Bowers, K.J. (2003) ‘Opportunity is in the eye of the beholder: The role ofpublicity in crime prevention’. Criminology and Public Policy,2(3), 201-228.Johnson, S.D., Bowers, K., and Hirschfield, A. (1997) ‘New Insights into the Spatial and TemporalDistribution of Repeat Victimisation’. The British Journal of Criminology, 37(2), 224-244.Johnson, S.D., Bernasco, W., Bowers, K.J., Elffers, H., Ratcliffe, J., Rengert, G. and Townsley,M.T. (2006) Space Time Patterns of Risk: A Cross National Assessment of Residential BurglaryVictimization. Submitted.Johnson, S.D. and Bowers, K.J. (2004) The burglary as clue to the future: the beginnings ofprospective hot-spotting. European Journal of Criminology, 1(2), 237-255.Johnson, S.D. and Bowers, K.J. (2004) ‘The stability of space-time clusters of burglary’. The BritishJournal of Criminology, 44(1), 55-65.Johnson, S.D., Bowers, K. and Pease, K. (2005) ‘Predicting the Future or Summarising the Past?’Crime Mapping as Anticipation. In Crime Science: New Approaches to Preventing and DetectingCrime, M. Smith and N. Tilley (Eds.) 145-163.71

Johnson, S.D., Summers, L., and Pease, K. (2006) Vehicle Crime: Communicating Patterns of Riskin Space and Time. Report to the Home Office.Knox, G. (1964). ‘Epidemiology of Childhood Leukaemia in Northumberland and Durham’. BritishJournal of Preventative and Social Medicine, 18, p17-24.Lamm Weisel, D. (2002) ‘Burglary of Single-Family Houses’. Problem-Oriented Guides for PoliceSeries, No.18. Washington: U.S. Department of Justice.Laycock, G. and Tilley, N. (1995) Policing and Neighborhood Watch: Strategic Issues. CrimeDetection and Prevention Series, No. 60. London: Home Office.Laycock, G. (1985) Property Marking: a deterrent to domestic burglary? Crime Prevention Unit, PaperNo. 3. London: Home Office.Lipsey, M.W. and Wilson, D.B. (2001) Practical Meta-Analysis. Thousand Oaks, CA: Sage.Lister, S., Wall, D. and Bryan,J. (2004) Evaluation of the Leeds Distraction Burglary Initiative. HomeOffice Online Report No. 44. London: Home Office.Marchant, P. (2005) ‘What Works? A Critical Note on the Evaluation of Crime Reduction Initiatives’.Crime Prevention and Community Safety: An International Journal, 7(2), 7-14.McLaughlin, L.M., Johnson, S.D., Bowers, K.J., Birks, D.J. and Pease, K. (2007) ‘PolicePerceptions of the Long and Short Term Spatial Distribution of Residential Burglary’. The InternationalJournal of Police Science and Management, in press.Nelson, D., Collins, L. and Gant, F. (2002) The stolen property market in the Australian CapitalTerritory. Canberra: Australian Institute of Criminology.North, B.V., Curtis, D. and Sham, P.C. (2002) ‘A note on the Calculation of Empirical P Values fromMonte Carlo Procedures’. American Journal of Human Genetics, 71/2, 439-441.Openshaw, S. (Ed) (1995) Census Users’ Handbook. Cambridge: GeoInformation International.Pease, K. (1998) Repeat Victimisation: Taking Stock. Preventing and Detecting Crime, Paper 98.London: Home Office.Polvi, N., Looman, T., Humphries, C. and Pease, K. (1991) ‘The Time Course of Repeat BurglaryVictimisation’, British Journal of Criminology, 31(4), 411-14Ratcliffe, J. H. (2002) ‘Aoristic signatures and the temporal analysis of high volume crime patterns’.Journal of Quantitative Criminology, 18 (1): 23-43.Ratcliffe, J. H. (2005) ‘Detecting spatial movement of intra-region crime patterns over time, Journal ofQuantitative Criminology’. 21(1): 103-123.Ratcliffe, J.H. (2006) Detecting the near repeat phenomenon in local and regional crime data. NIJgrant.Rengert, G. and Wasilchick, J. (2000) Suburban Burglary: A Tale of Two Suburbs. Springfield, IL:Charles Thomas.Riley, D. (1980) ‘An Evaluation of a Campaign to Reduce Vandalism’. In Designing Out Crime, Clarke,R.V. & Mayhew, P. (eds.). London: HMSO.Shadish, W.R., Cook, T.D., and Campbell, D.T. (2002) Experimental and Quasi-Experimental Designs for Generalized Causal Inference. Boston: Houghton Mifflin Company.72

Simpson, E.H. (1959) ‘The Interpretation of Interaction in Contingency Tables’. Journal of the RoyalStatistical Society, Series B, 13, 238-241.Smith, M.J., Clarke, R.V., and Pease, K. (2002) ‘Anticipatory Benefit in Crime Prevention’. InAnalysis for Crime Prevention: Crime Prevention Studies 13, Tilley, N. (ed.) Criminal Justice Press.Spelman, W. (1994) Crime Incapacitation. New York: Plenum.Stockdale, J. E. and Gresham, P.J. (1995) Combating Burglary: An Evaluation of Three Strategies.Crime Detection & Prevention Series, Paer 59. London: Home Office.Sunder, N. and Birks, D.J. (2004) ‘Measuring Incidence, Prevalence and Concentration: Implicationsfor Policing’. Crime Prevention and Community Safety: An International Journal, in press.Sutton, M. (1998) Handling stolen goods and theft: a market reduction approach. Home OfficeResearch Study, No. 178. London: Home Office.Tilley, N. (1993) ‘After Kirkholt – Theory, Method and Results of Replication Evaluations’. PoliceResearch Group Crime Prevention Unit series Paper 47. London: Home Office.Tilley, N. and Hopkins, M. (1998) Business as Usual: An Evaluation of the Small Business and CrimeInitiative. Police Research Series, Paper No. 95. London: Home Office Police Department.Tilley, N and Webb, J. (1994) Burglary Reduction: Findings from Safer Cities Schemes. CrimePrevention Unit Series, Paper No. 51. London: Home Office Police Department.Townsley, M. Homel, R. and Chaseling, J. (2003) ‘Infectious Burglaries: A Test of the Near RepeatHypothesis’. British Journal of Criminology, 43, 615-633Tseloni, A. and Pease, K. (2004) Population Inequality: The Case of Repeat Victimisation. Insubmission.Wagner, A. (1997) ‘A Study of Traffic Pattern Modifications in an Urban Crime Prevention Program’.Journal of Criminal Justice 25(1):19-30.Welsh, B. C. and Farrington, D.P. (2002) Crime Prevention Effects of Closed Circuit Television: ASystematic Review. Home Office Research Study, 252. London: Home Office.White, G. (1990) ‘Neighbourhood Permeability and Burglary Rates’. Justice Quarterly 7(1):57-67.73

Appendix 1. The information technology nexusMuch of this report concentrates on the principles of prospective mapping and police officers’reactions to it. If systems such as Promap are to become commonplace in an operational policingcontext, of equal importance is the practicality of producing the maps on a routine basis.The integration of novel information technology based analysis tools into an existing IT framework canproduce several challenges. In this section, possible methods of delivering them are discussed, somekey issues and problems associated with the development and integration of them identified, and anumber of solutions (both short- and long-term) to these suggested.In order to produce prospective maps, as a minimum crime data relating to the timing and location ofprevious events are required and must be extracted from existing crime recording systems. Thismeans that the ability to retrieve, process and analyse these data in a timely fashion is of keyimportance to the approach.In the main, as any new software must in effect be grafted onto force IT systems, there are two mainapproaches to implementing new analytic software products.1. New software may be integrated into existing systems. This solution would mean that datacould be directly accessed by the new software from an internal structure/datawarehouse, process it and produce output within existing analysis systems. This option ispresented schematically as Figure A1.1. It would require the least intervention on the partof the analyst and is the preferred option. However, there is typically a reluctance from ITstaff regarding this approach. The reason for this is simple. Where software is notproperly tested, it is possible that the computer program code could have a detrimentalaffect on existing systems, slowing them down or much worse.Consequently, this option requires considerably more effort in terms of systemdevelopment and testing. A programme of formal specification would be required. At leastthree stages are desirable. Program validation would ensure that the software providedsufficient functionality to serve its purpose. Program verification on the other hand wouldbe required to check that the software is implemented correctly. Additionally, an extensivetesting regime would be needed to demonstrate that the system would not have anyunwanted side effects on other elements of the IT infrastructure. For even a relativelyinexperienced programmer the latter is unlikely but testing is an important step.In addition, it is worth noting that any direct integration of a new application into anexisting information technology infrastructure is understandably a complex process. Froma software implementation perspective the majority of police forces currently implementunique data infrastructures. This means that all attempts at integration must be tailored tointerface with specific systems, Subsequently successful products are not immediatelytransferable between systems. This issue is currently being addressed to some extent bythe IMPACT programme, which aims: “To deliver an effective integrated national, regionaland local information-sharing and intelligence capability, which will improve the ability ofthe police and partner agencies to proactively use information for intelligence purposes toprevent crime, bring offenders to justice, safeguard children and vulnerable people andfurther professionalise the investigation process”.74

Figure A1.1: Internal application utilising existing GIS for visualisation2. A stand-alone application may be developed that can process crime data exported fromexisting IT systems. The major disadvantage of this approach is that data have to first beextracted, stored on some form of portable media and then imported to the externalapplication. This takes time, particularly where the maximum rate of data transfer possiblefor the portable media used is slow. The use of Universal Serial Bus hard disk dives (USBHDD) or Firewire connections can increase the celerity of data transfer, but suchconnections are typically disabled on police IT systems for reasons of security.Once the data have been processed they can be relayed to the user in one of two ways.a) As shown in Figure A1.2, data may be imported into an existing GeographicalInformation System and used to produce maps. A disadvantage with thisapproach is that further time is required to import and manipulate the data, and toadd any required images of the urban backcloth.b) Alternatively, a GIS visualisation tool may be implemented within the stand-aloneapplication to allow maps to be generated and interrogated within it (see FigureA1.3). This has the distinct advantage that the process of visualising the data canbe automated and images may be produced to a standard (but modifiable)template.75

Figure A1.2: Stand-alone application utilising existing GISFor reasons already discussed, option 2b was used in the current project. That is, a stand-aloneapplication was developed which accepts data from a predefined export procedure installed on apersonal computer with access to a live crime recording system. The application runs on a laptop,providing all the processing and visualisation elements. As the visualisation component of the systemwas specifically designed for this purpose, the relevant images of the urban backcloth areautomatically selected by the application at the correct level of resolution for the map displayed.76

Figure A1.3: Standalone application with integrated GISWhilst this was the optimal solution for the current project, option 1 discussed above is the desiredapproach. However, whatever option is used in future projects, a number of recommendations withrespect to police IT systems and the recording of data that may be realised in the short and longerterm are presented below.Recommendations that may be realised in the short term• Crime data should be recorded in an accurate and timely manner. Georgaphical grid coordinatesshould be accurate to a resolution of one metre and available for analysis within 24hours or less.• All force IT systems should have standard applications that enable data to be exported in astandardised format, such as the comma separated (.csv) format which is compatible withmost commercial software, and programming languages.• Physical access to data should be possible, either using a CD burner or (for faster datatransfer) USB HDD. Security measures would, of course, be required to restrict access to thedata. Different approaches to encrypting data should be explored.77

Recommendations that may be realised in the longer term• Force IT systems should be developed with standardised interaction architectures, whichallow the incremental addition of analysis elements to both crime recording (such as datacleaning) and analysis systems (such as Promap).• Standardise recording and IT architectures across different forces, regions and organisations.This would facilitate the use of the same applications across police forces and thus eliminatethe need for bespoke applications. This would also facilitate analysis of cross borderoffending.78

Appendix 2. Prospective Mapping SurveyPlease read carefully:The UCL Jill Dando Institute of Crime Science has been commissioned by the Home Office and theGovernment Office for the East Midlands (GOEM) to undertake a research project that has beenpiloted in Derbyshire ‘A’ Division from August 2005 to February 2006. It is essential with a project ofthis kind that an appropriate evaluation is carried out to understand and assess exactly whathappened and to identify any factors that particularly facilitated or impeded implementation.This survey has been designed by the research team to gain an understanding of how much you knowabout the pilot and how useful you feel the maps were in ‘A’ Division, and more specifically, yoursection.Please note that this is an anonymous survey. Your individual answers will not be seen by anyoneother than our research team and we will not be able to identify you by the answers that you give, norwould we wish to do so.Thank you for taking the time to fill out this questionnaire.Please turn over to begin79

Section 1: knowledge and understanding of prospective mapping1a) Have you heard of the prospective mapping pilot that is taking place in ‘A’ Division?(please tick)Yes .......................................................................................... (if yes, please proceedto 1b)No............................................................................................ (if no, please proceedto Section 3)1b) In your own words, please briefly say what you understand prospectivemapping to be?1c) Have you seen any prospective maps over the last six months?Yes .......................................................................................... (if yes, please proceedto 1d)No............................................................................................ (if no, please proceedto Section 3)1d) Were these maps used by yourself or your supervising officer for targeted policeactivity (e.g. directed patrols, targeting offenders)?Yes ......................................................................................... (if yes, please proceedto 1e)No............................................................................................ (if no, please proceedto 1g)Don’t know .............................................................................. (if don’t know, pleaseproceed to 1g)1e) How often do you (or your supervising officer) use prospective map(s) for targetedpolice activity?You supervising officerMore than twice a weekTwice a weekOnce a weekOnce every two weeksOnce every three weeksOnce every monthLess than once a month80

1f) What tactics have you employed, or been involved in, in response to the maps?Foot patrol around hotspotsDrive through patrols in hotspotsTarget-hardening (e.g. locks, alarm systems)Repeat victimisation strategiesTargeting known offendersRedeployable CCTVPublicity campaignsStreet closuresAlleygatingEmployedInvolved inOther, please specify: ………………………………………………………………………………………………………………………………………………………………………………………………………………………………….........................1g) How easy do you feel the maps are to interpret?Easy ........................................................................................Fairly easy...............................................................................Fairly difficult ...........................................................................Difficult ....................................................................................I don’t understand the maps ...................................................Extra Comments (please outline any extra comments you have about your knowledgeand understanding of prospective maps and the tactical options that you haveemployed)Section 2: usefulness of the maps2a) How useful do you feel the maps are in your day-to-day work?Very useful ..............................................................................Somewhat useful ...................................................................Not very useful........................................................................Not useful ...............................................................................2b) Have the maps ever identified risky areas that you would not otherwise haveconsidered as being risky?Often .......................................................................................Sometimes ..............................................................................Never.......................................................................................Yes, but I trusted my own judgement rather than the maps ..81

Extra Comments (please outline any extra comments you may have about theusefulness of prospective maps, how they might be improved etc.)Section 3: personal details(Please remember that this survey is both confidential and anonymous and the information you givewill only be used by the research team to help put your answers into context.)4a) What is your sex?Male ........................................................................................Female ....................................................................................4b) What is your current rank/position?Police Constable.....................................................................Sergeant .................................................................................Inspector .................................................................................Other (please state)................................................................4c) How long have been in your current rank?Less than 1 year ....................................................................1-5 years .................................................................................More than 5 years ...................................................................4d) What Section do you work in?Alfreton....................................................................................Belper......................................................................................Ilkeston....................................................................................Long Eaton..............................................................................Ripley ......................................................................................82

Appendix 3. Detailed evaluation methodologyThe purpose of this appendix is to describe the evaluation techniques used that were not discussed inthe main body of the report. Some of these are novel and have a potentially wider application thanthis project and are thus discussed so that others might use or improve upon them. This section of thereport is intended to be self-contained and so there is some duplication with the text presented in themain report, although this repetition is minimal. The methods discussed consider changes observedover time, in space and in space and time, in that order.Analyses of change over timeWhich statistical method is most appropriate for establishing the statistical significance of changes inlevels of crime in a single area over time is the matter of some debate. A number of approaches exist.First are those that consider the overall difference in the volume of crime before and after intervention.The basic approach is to compare the change in the volume of crime for a particular unit of time (say12 months) before and after intervention in both an action and comparison area. If a reduction isobserved in the action but not comparator, or the reduction in the former exceeds that in the latter thena positive inference may be drawn. To determine whether the difference in the change between thetwo areas is significant, a measure of effect size and the associated standard error is derived.There are a number of approaches to computing effect sizes for single-case designs (Allison andGorman, 1993; Lipsey and Wilson, 2001). Here attention will be given to two techniques: one usedwithin the criminological literature and elsewhere, the other developed within the field of psychologyfor the analysis of behavioural change, but (variants) also used more widely within other fields ofinvestigation, such as economics.The simplest approach is to compute an odds ratio, which simply compares the change in theintervention and comparison areas before and after intervention. An odds ratio of one indicates thatthe changes in the two areas were commensurate, suggesting no impact of the scheme. An oddsratio of greater (less) than one suggests a reduction (increase) in the intervention area relative to thechange observed in the comparison area. The statistical significance of the odds ratio can also becomputed (see Lipsey and Wilson, 2001) by estimating the standard error of the value derived. Thistechnique, which is readily interpretable, has been frequently used in research concerned with whatworks in reducing crime (for examples, see Welsh and Farrington, 2006; Gill and Spriggs, 2005), but isnot without its critics for analyses conducted at the small area level (for which fluctuations over timemay occur even in the absence of intervention: Marchant, 2005). However, the problems articulatedabout this approach are likely to be less problematic for analyses conducted at the BCU level, forwhich the variation over time is likely to be relatively stable. Thus, the approach is used here not leastbecause it provides a simple assessment of how things changed in the pilot area compared to thecomparator.Two approaches are here used to compute the standard errors (since these are critical in determiningthe significance of the effect-size derived), one used by Farrington and colleagues (see Welsh andFarrington, 2006), the other by Gill and Spriggs (2005). 10 However, both approaches converged onsimilar estimates and hence only the former are presented.An alternative to using data which has been aggregated for two periods of time (before and afterintervention) is the analysis of time-series data. For this approach, data for a number of intervals areinstead analysed. This allows more complex patterns in the data to be identified and considered inthe analysis. For example, time-series analysis can help control for what is known as serialdependence in the data: that is, to control for the fact that the residual error for an observation at onetime point is likely to be highly related to that for an adjacent time point. If such dependence existswithin the data then failing to correct for it can increase the likelihood of Type I statistical error – thelikelihood of incorrectly rejecting the null hypothesis.10 Gill and Spriggs (2005) use a slightly different approach to calculate the standard by considering monthly fluctuation in thevolume of crime to reduce a problem known as over-dispersion.83

In relation to the evaluation of interventions, the method used is known as an interrupted time-seriesdesign (Shadish, Campbell and Cook, 2002). The rationale for the approach in the current contextwould be that if an intervention has an impact on the crime rate of an area then followingimplementation the trend in the crime rate should change. This can occur in at least two ways. First,the mean level of the series may change, or there may be a change in the slope of the time-seriesfollowing the inception of a scheme. Thus, following intervention the average monthly crime rate mayfall by an average amount (say 20 crimes per month), or there may be a downwards trend with thecrime rate falling by an incremental amount each month.The basic method used is to first analyse the data for the pre-intervention series. The purpose of sodoing is to derive a set of parameters that describe monthly (or some other interval of time) changes inthe crime rate before implementation. These parameters include any trend in the series (linear orotherwise), the intercept of the series (the value at time zero), and the extent of serial-correlation in thedata. Other parameters may also be modelled but are not discussed here for parsimony. Once theseparameters have been estimated they can be used to see how well they describe the post-interventiontime-series. Moreover, and importantly, the aim of the analysis is to see if the timing of intervention,modelled as a binary variable (or a continuous variable if data on the intensity of implementation areavailable) explains a significant amount of the variation in the time-series that is not already explainedby the parameters estimated in the earlier steps of the analysis.In many fields of investigation, data are often available for long time-series (say 30-50 months) beforeand after intervention and this allows reliable analyses to be conducted. In the field of crime reduction,the length of the series is typically shorter. In the current evaluation, although a long series wasavailable for the pre-intervention series, the post-intervention was fairly short, at only seven months.Even where a time-series is short Shadish et al. (2002) recommend the use of time-series analysis,even if the analysis undertaken is simply visual inspection of the trend.However, time-series approaches have also been adopted in other fields of investigation for which thelength of the series is frequently much shorter. For example, in reviewing the methods available forthe calculation of effect sizes for single-case designs where a control group is unavailable, Allison &Gorman (1993) propose a method for the analysis of shorter time-series. The rationale behind theapproach is that in the absence of intervention, the time-series before intervention can be used topredict that afterwards. Their approach uses ordinary least squares (OLS) regression to diagnose thegeneral trend before intervention, and the resulting regression equation is applied to the data postintervention.Their formulation allows both changes in intercept and slope to be identified. Formally:Y = b 0 + b 1 X + b 2 X(t-n a ) + eWhere,Y is the residual error from the initial OLS modelb 0 is the estimate of the interceptb 1 is the estimate of the change in intercept associated with the interventionb 2 is the estimate of any change in slope associated with the interventiont is the time pointe is the error termWhere data concerned with changes in a control group are unavailable, this approach would help rulea number of threats to internal validity; that is, factors other than an intervention that might explain thepattern observed. Such threats include (for example) regression to the mean and history. Regressionto the mean may occur where an area is selected for intervention on the basis of an extreme preinterventioncrime rate, and where the rate is extreme not only compared to other areas but relative toitself at other times. The problem is that the observed elevation in the crime rate may be explained bytemporary phenomena. Thus, even in the absence of intervention the crime rate would soon regressback to the level typical for that area. With a suitably long time-series of data, such effects can beidentified and modelled. Similarly, history occurs where there is a downwards trend in the crime rate84

prior to intervention, or specific events unrelated to the intervention occur that might impact upon theincidence of crime.However, one limitation of the type of analysis discussed above is that it is insensitive to more generalchanges that might occur. Consider that there may be changes in policy that would have a genericeffect on the rate of burglary. For example, a county-wide policy introduced on a particular date wouldbe expected to impact upon burglary across the entire area. Unless such a policy were identified itcould not be modelled using the type of analytic design discussed above. Changes in recordingpractices, such as those introduced as part of the National Crime Recording Standard, may also affectthe time-series.Thus, a variant of the above approach for which changes in a comparison area are used to de-trendthe time-series in the first step of the analysis are here used to increase the validity of the analysis. Bydoing so, this means that any changes in the action area that may be explained by factors such asthose already discussed can be estimated and removed from the time-series.For the current evaluation a fairly long pre-intervention time series facilitated a reliable diagnosticanalysis of the pre-intervention time-series. However, many researchers recommend that wherepossible it is wise to employ a range of statistical approaches to the analysis of data. The advantageof so doing is that where the results of the analysis converge, one can be more confident thatconclusions drawn are not based on the specific biases of a particular statistical procedure. For thisreason, in this report analyses are conducted using both the odds ratio approach discussed above anda variant of the approach to the analysis of short time-series proposed by Allison & Gorman (1993).Odds ratio analysisTo calculate the odds ratios, the count of crime for the seven months before and during the pilot werecontrasted. The standard errors were computed in the usual way (see Lipsey and Wilson, 2001) aswell as using monthly variation as suggested by Gill and Spriggs (2005). Table A3.1 shows the countof burglary before and during the pilot along with the OR, and associated confidence intervals and z-score. The confidence intervals shown, calculated using the traditional approach indicate the upperand lower estimates of the odds ratios. These suggest that the true odds ratio lies somewherebetween 0.99 and 1.34. The z-score provides an indication of the likely statistical significance of theOR. For a two-tailed test, the z-score is required to exceed 1.96. On the basis of these results, theanalysis would suggest that the reduction in burglary observed in ‘A’ Division exceeded that in thecomparison area, but that the trend was marginally non-significant.Table A3.1: Change in the volume of burglary and odds-ratio statisticsBefore After GrosschangeOddsratioConfidenceintervalsz-scorePilot 663 554 109 1.15 0.99-1.34 1.80Comparisonarea684 659 2585

Time-series analysisFor the purposes of illustration, Figure 6.1 (in the main body of the report) shows a time-series graphof the change in the count of burglary in the action and control areas for a period of two years beforeintervention up until the end of the evaluation period. As a first step, an OLS analysis, with themonthly count of burglary in the action area for the period prior to implementation was regressedagainst that in the comparator. Also included in the model was an (independent) variable whichindicated the time point (see above). This analysis indicated that changes in the comparison area, ‘C’Division, were significantly associated with those in the action area (ß=0.33, ß(s.e)=0.09, t(51)=3.90,p

explained by those in the control area, there was no additional temporal trend (upwards ordownwards) in the changes observed (ß=-0.01, ß(s.e)=0.01, t(51)=-1.14, p=0.26).Again, a corrolellogram confirmed that the data were serially correlated, conforming to an AR1 model,and thus this model was used to see if the reduction observed was coincident with the timing of thepilot. The results of the analysis confirmed that a significant amount of the variance in the weeklyburglary count in the Division was explained by serial autocorrelation (ß=0.74, ß(s.e)=0.04,t(245)=17.22, p

Figure A3.1: Changes in the spatial distribution of risk following the introduction of the pilotIt is important to bear in mind that this is a purely visual and descriptive analysis. Moreover, thepatterns are relative and do not necessarily indicate where the risk of burglary was greatest, but justwhere it changed the most. Nevertheless, what Figure A3.1 does illustrate is that risk moves evenwhen considered over fairly lengthy intervals of time, such as seven months. This mobility of risk isone of the underpinning justifications of Promap.With respect to statistical significance, an analytic approach worthy of discussion has been developedby Ratcliffe (2005). The aim of this is to detect changes in the spatial placement of crime over timewithin a particular area, such as a BCU. Using a Monte-Carlo simulation, the spatial patternsobserved can be compared against the null hypothesis that any changes observed could haveoccurred on the basis of chance. Consequently, the test developed provides an inferential statisticaltest of the visual analysis presented in Figure A3.1.88

A still further approach to exploring the stability in burglary risk is to compute the correlation betweenthe count of burglary across a series of areas encapsulated by the BCU before and during the pilot. Alarge correlation would suggest that the distribution of risk was stable, a small coefficient that it wasfluid. To do this, areas at different levels of geographical resolution could be used. For example, onecould examine the correlation at the level of a town or for a grid of smaller regular sized cells.Obviously, the larger the area considered, the more stable the patterns would be expected to be.Thus, analyses were conducted at two levels of geographic resolution, first at the census output arealevel, and second for a series of 1kmX1km cells. There are a total of 29 census output areas in ‘A’Division and at this level of resolution the rank-order correlation of .89 (p

Figure A3.2: Lorenz curves showing the distribution of burglary riskCumulative percentage of burglary100908070605040302010Cumulative % afterCumulative % before00 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95100Cumulative percentage of 1km areasThe general pattern for both periods of time confirms what the authors already knew, that burglary ishighly concentrated in space. It also shows that there was little change in the spatial concentration ofburglary following the start of the pilot. This is slightly disappointing. With fuller implementation, itwould become worthwhile to compare Lorenz curves for pilot and comparison areas, since there maybe a general tendency for crimes as they become rarer also to have a very different distribution.Changes in patterns in space and timeThe burglary pattern which gives prospective mapping its superiority over retrospective mapping is itsspatio-temporal clustering. The traditional definition of a geographical hotspot is an area for whichcrime is particularly concentrated in space for a period of time such as one year or six months. Thus,hotspots are essentially the accumulation of pairs of crimes which occur close to each other within thetime period of interest. In consequence, the analysis of hotspots does not consider the timing 12 ofevents, other than requiring them to have occurred within the interval of interest, say over the last sixmonths. The analyses so far presented have looked at distributions in time and space separately. Analternative and more sensitive approach to the examination of spatio-temporal patterns of crime wouldbe to identify clusters or series of event pairs that occur not only close in space but also in time. Puttoo simply, one would expect prospective mapping to reduce the length of series of burglaries close intime and space.The identification of such clusters and their length would have useful implications for crime reductionin at least two ways. First, this would provide the police with an idea of the tempo of localisedincreases in burglary risk. For example, if the longest clusters that could be identified consisted ofonly two events, this would suggest that the elevated risk in an area had a fairly slow tempo, whereasif clusters of ten events were routinely identified this would suggest a higher frequency of space-timeclustering. This form of analysis chimes with the Promap approach. It is also novel. The reader isinvited to think of what follows in this light, namely as an exploratory approach which no doubt needsrefinement but is more suited to the technique it aspires to understand and evaluate. This type ofanalysis is useful for evaluation research by enabling the detection of qualitative changes in spatiotemporalpatterns of crime. It is perhaps easier to illustrate this point by starting with a special case of12 By timing, the authors mean the time between events rather than the regularity of the time of day that crimes occur. The latterhas been usefully explored by Ratcliffe (2002).90

space-time clustering: repeat victimisation. Considering evaluations of area-based interventions, it ispossible to not only measure changes in burglary incidence (the rate per 1,000 households) followingan intervention but also how concentrated on individual properties crime is. For example, for anintervention aimed at reducing repeat victimisation, simply examining the change in the incidence ofburglary may be insufficient to detect the full impact of a scheme, or the potential crime reductionmechanisms through which any change was realised. Instead, a more sensitive analysis (seeForrester et al., 1990) would be to examine the rate of repeat victimisation. A reduction in this type ofvictimisation would suggest an impact of the scheme, although a reduction in incidence would berequired to demonstrate that target-switch displacement did not occur. On the other hand, a reductionin incidence within an area that was unaccompanied by a reduction in repeat victimisation wouldprovide less convincing evidence that any reduction could reasonably be attributed to the scheme.In the same way, where strategies are employed to suppress emerging or enduring hotspots of crime,as was the case in the current pilot, one desired outcome of the scheme would be to truncate thelength of space-time clusters of burglary. That is, if resources are directed to the right places at theright times in a way that anticipates where the next event in a cluster is most likely to occur, then itwould be hoped that emerging spatio-temporal clusters of crime would be targeted before they have achance to propagate.To examine this issue, software was developed to identify series of events that occurred close in bothspace and time, ranging from two events (pairs) onwards. The approach allowed the frequency ofseries of different (k-event) lengths (e.g. pairs, triples, quads and so on) to be summarised andcompared for the periods before and during the pilot.The identification and summary of series of events can be done in a variety of ways. Here, for everyburglary (the reference) event, any antecedent burglary that occurred within a critical distance andtime of it was identified and added to the series. Using this approach only burglaries that occurredwithin the critical time and distance of the reference event could be identified. This precludes theidentification of other burglaries that might be linked to the others in a cluster. This problem isillustrated conceptually in Figure A3.3. In the top and bottom of Figure 6.8, there are four burglariesthat occurred within a few days of each other. In the first chain, all burglaries occur within the criticalspatial distance of the reference event (the leftmost event) and thus one identifies a chain of fourburglaries. For the second series, for the same reference event one would identify a chain of onlythree burglaries as the final event occurred further away from the reference event than the criticaldistance. However, it occurred within the critical distance of other events in the chain and henceshould be included in the series.Figure A3.3: An illustration of a triple (bottom) and quad chain (top)Critical distanceAn alternative approach would be to include in a cluster all burglaries that are within the criticaldistance and time of one or more events already identified as part of that series. Using this approach,both series shown in Figure A3.3 would be classified as being a four event series. In what follows thecritical distance and time used were 400m and one-week, respectively. Other definitions could beused, and other analyses (not shown) using different definitions revealed a similar pattern of results.91

To recapitulate and elaborate, the approach taken may be summarised as shown below:1. The burglary data were sorted in date order.2. For each period of time (before and during the pilot), the previous week’s events were used asa historic buffer period to allow all events that occurred within seven days of another to beidentified.3. Each burglary event was then considered in sequential order.a. For the first event considered, only the historic data were searched to see if earlierburglaries occurred within the critical thresholds of the first event; where they did theywere added to the series for that reference event.b. For the second event considered, all of the historic data (including the first referenceevent) were searched, and events that belonged to a series identified as per step a.c. Once all crimes had been considered, for every event it was possible to calculate themaximum length of the chain for which that event was the terminal event and howmany chains of every length were identified.Using this approach, it was possible to answer the question, ‘when a burglary occurs how manycrimes previously occurred nearby in space and time?’ It is, of course, possible that any cluster soidentified could be part of a longer cluster which included events that subsequently occurred.Expressed a slightly different way, clusters identified in this way may overlap with one another withone cluster including some events of another – although for any particular series length no twoclusters would include exactly the same events – there would be at least one different crime in thelonger chain (the shorter being a subset of the other).Having summarised the data in this way, it was possible to calculate what proportion of burglarieswere the terminal event for different lengths of space-time series. The results, shown as Figure A3.4,illustrate that around one-third of events occurred within 400m and one week of at least oneantecedent both before and during the pilot. Around 15 per cent of events occurred nearby and withinone-week of at least two others. For the two intervals of time considered, the longest series identifiedconsisted of 12 burglaries that occurred close to each other in both space and time. Before discussingthe differences for the two periods of time, it is important to note the implication of this finding, which isthat it clearly illustrates the flux of crime. If burglaries occurred in the same places all the time, eitherconsiderably longer series would have been identified, or a higher proportion of events would belongto longer series. This emphasises the need for a dynamic predictive capability for the deployment ofcrime reductive resources. Simply targeting the same areas over time is unlikely to direct resources tothe right places at the right times.92

Figure A3.4: The proportion of events belonging to different k-event series before and duringthe pilot40% of events that are the last in a k- event (ormore) series3530252015105AfterBefore02 3 4 5 6 7 8 9 10 11 12Chain length (k -events)Considering the patterns for the periods before and during the pilot, there are some differences. First,the longest series (12 events, of which there were three examples) identified occurred in the periodbefore intervention. More generally, longer series were identified in the period before the pilot, thanduring it. Conversely, during the pilot there was a slightly higher proportion of shorter series (two orthree events). On the face of it, this finding is in line with what would be expected if the pilot had beensuccessful. However, the reader should be mindful of two things.a. For the period during the pilot, a smaller number of burglaries were committed, whichmay have affected the pattern of results. For what it is worth, sampling only the samenumber of events for the period before (the first 554 events) as that after generatedexactly the same pattern of results. Consequently, differences in sample sizes wouldnot appear to explain the results observed.b. Perhaps more importantly, the effect observed was very subtle. This is notunexpected given that implementation, or rather the use of the system, was at bestonly minimally realised. With such a small effect size, and under the circumstances ofpartial implementation, it is inappropriate to conduct inferential statistical tests. 13Thus, this finding is interpreted as being inconclusive, but the analytic approachderived suggested a useful method for future research concerned with the stability ofclusters of crime.13 Such a test would involve comparing the observed distribution of k-event series with what would be expected if the timing andlocation of events were random. Thus, a variant of the Monte-Carlo approach used elsewhere in this report would seemappropriate.93

Concluding comments on methodThe techniques described illustrate a number of ways in which particular ‘signatures’ that mightbespeak mechanism may be sought. Only by conducting such analyses is it possible to differentiatebetween the likely contributions of different plausible mechanisms of crime reduction. Unfortunately,methods of this kind are in their infancy and hence there is a need to develop and disseminate thosethat might have wider application.94

Appendix 4. Promap graphical user interface and anillustration (step by step) of how the system is usedThis appendix provides further detail concerning the Promap interface, how it is used, and the outputgenerated. The maps produced show anticipated risks at various levels of geographic resolution, thefinest being at the household level. However, for the purposes of anonymity, all of the maps shownhere are at a fairly coarse level of resolution and the predictions shown are fictitious rather thanreflecting the actual distribution of crime risk.For the purposes of elaboration, the notes that follow describe in a step-by-step fashion how thesoftware is used. These are a version of the instructions provided to the crime analysts.Step 1:Load the Promap application.Figure A5.1: The Promap applicationStep 2:If you wish to produce a prospective map for the current date, select the ‘today’ radio button (seeFigure A5.1). If you wish to change the test date select ‘other’ and use the date selection box. If youare producing a map for a specific shift – enable the shift analysis mode with the tick box and then usethe radio buttons to select which shift the map should be tailored for (see Figure A5.2).95

Figure A5.2: An enlargement of the shift analysis optionsStep 3:Once you are happy with your selection, click the 'Generate PROMAP' button, this will perform theanalysis and generate the prospective maps required. While these calculations take place a‘processing data’ message will remain in the foreground of the map window.Step 4:When analysis is complete the processing message will disappear and the prospective map of yourarea will become visible (as below). The key in the bottom left of the map indicates the fraction of thearea identified. This ranges from the five per cent (shaded yellow), two per cent (shaded orange) andone per cent (shaded blue) of the total area predicted to be most at risk of victimisation. Thus,according to the forecast, the blue area is that which should be prioritised first.Figure A5.3: An example of fictitious prospective map© Crown Copyright. All rights reserved. Derbyshire Constabulary 100021015.96

Step 5:Once the prospective map has been generated, navigational controls in the bottom right of the window(shown in Figure A5.4) allow further user interrogation of the map. Brief descriptions of the functionsare shown below.Figure A5.4: Map navigational options• The control allows the user to navigate the map at the current level of magnification,moving the map in combinations of East/West and North/South.• The button zooms into the map window providing more detail for a specific area. Thiscan be done by clicking at a particular point on the map or dragging a box around the area toview. Promap will dynamically switch between the relevant Ordnance Survey map layers forthe level of zoom selected.• The magnifying glass zooms out of the map window.Figure A5.5: Prospective map magnified to neighbourhood level.© Crown Copyright. All rights reserved. Derbyshire Constabulary 100021015.97

Figure A5.6: Prospective map magnified to street level© Crown Copyright. All rights reserved. Derbyshire Constabulary 100021015.Figure A5.7: Prospective map magnified to household level. © Crown Copyright. All rights reserved. Derbyshire Constabulary 100021015.98

Using the navigational tools, it is possible to interrogate the map at the level of resolution desired.Such analysis may be combined with other intelligence to help further prioritise resource allocationwithin the areas identified and to refine tactical options.The 'export view as image' button does exactly what it says – it exports the current map window viewas a gif image for inclusion in documents or briefings. By default all captured images can be found in‘C:\promap\exported_images\’ and are given a name based upon the area in question, and the typeand date of analysis performed.99

Produced by the Research Development and Statistics Directorate, HomeOfficeThis document is available only in Adobe Portable Document Format (PDF)through the RDS websitehttp://www.homeoffice.gov.uk/rdsEmail: public.enquiries@homeoffice.gsi.gov.ukISBN: 978 1 84726 532 6© Crown copyright 2007